Electroencephalography and clinical Neuropl~vsiologv, 1986, 6 4 : 4 8 3 - 4 9 2 Elsevier Scientific Publishers Ireland, Ltd.
483
Clinical Section C O M P U T E R I Z E D EEG SPECTRAL A N A L Y S I S IN ELDERLY N O R M A L , D E M E N T E D AND DEPRESSED SUBJECTS R I C H A R D P. B R E N N E R *'**, R I C H A R D F. U L R I C H *, D U A N E G. SPIKER *, R O B E R T J. SCLABASSI *** C H A R L E S F. R E Y N O L D S , III *'**, R O B E R T S. M A R I N * and F R A N C O I S BOLLER *'** Departments" of * Psychiato', ** Neurology, *** Neurological Surgery and Electrical Engineering, Universi(v of Pittsburgh, School of Medicine, Pittsburgh, PA 15213 (U.S.A.) (Accepted for publication: April 29, 1986)
S u m m a D, Computerized spectral analysis of the EEG was performed in 35 patients with Alzheimer's disease and compared to patients with major depression (23) and healthy elderly controls (61). Compared to controls, demented patients had a significant increase in the theta and alpha~ bandwidths as well as an increased theta-beta difference. The parasagittal mean frequency, beta I and beta 2 activity were significantly decreased. Depressed patients differed from demented patients, particularly at the lower end of the spectrum, having significantly less delta and theta activity. Like the demented group, depressed patients also had a decreased parasagittal mean frequency, beta l and beta 2 when compared to controls. In demented patients, there was a high correlation between several spectral parameters (parasagittal mean frequency, delta and theta activity, and the theta-beta difference) and the Folstein score. E E G measures used for discriminant analysis were more accurate in identifying demented patients who had lower Folstein scores. Keywords: spectral analysis - dementia - depression - health)' elder(v
Interest has developed in computerized frequency analysis of EEG, in part, because the overlap in visual EEG findings in the normal, demented and depressed elderly populations make it difficult to categorize reliably the individual patient (Goodin 1985). Computerized analysis provides quantitative data and has been utilized in the evaluation of normal aging (Roubicek 1977; Duffy et al. 1984b; Dustman et al. 1985), dementia (Gerson et al. 1976; O'Conner et al. 1979; Canter et al. 1982; Coben et al. 1983; Prichep et al. 1983; Duffy et al. 1984a; Coben et al. 1985; Penttila et al. 1985; Visser et al. 1985), and depression (O'Conner et al. 1979; Visser et al. 1985). The diagnostic utility of this technique, however, remains uncertain. Therefore, to study group differences, as well as to identify EEG discriminant functions, and to assess the efficacy of spectral Address correspondence and reprint requests to: Richard P. Brenner, M.D., Western Psychiatric Institute and Clinic, E E G Laboratory, 3811 O ' H a r a Street, Pittsburgh, PA 15213, U.S.A.
analysis, we compared the results of spectral frequency analysis of EEG in 3 groups of elderly individuals: (1) healthy controls, (2) non-depressed patients with Alzheimer's disease (AD), and (3) non-demented patients with major depression.
Methods
(1) Subjects (Table I) Group 1. Thirty-five patients who met criteria for probable AD (McKhann et al. 1984) participated in the study. No patient had a history of major psychiatric illness and all had a Hamilton depression rating (Hamilton 1967) of 10 or less on the first 17 items of the scale (single rater). This test is an aid to the clinical examination and helps quantify severity of depression; non-depressed individuals usually have scores of 10 or less. Each patient underwent general medical, neurologic and psychiatric evaluations. Laboratory studies, including a CBC, chemistry screen, thyroid function
0013-4649/86/$03.50 cc 1986 Elsevier Scientific Publishers Ireland, Ltd.
484
R.P. BRENNER ET AL.
TABLE I Sample characteristics: mean ± S.D. n Gender (F/M) Age Age range Education (years) Folstein range Hamilton range
Controls
Demented Depressed
61 37/24 66.9 (6.8) 51-82 14.5 (3.3) 29.3 (0.8) 28-30 1.6 (1.5) 0-6
35 23/12 67.4 (8.9) 51-89 12.4 (2.4) 19.0 (5.3) 9-29 4.8 (2.7) 0-10
23 15/8 69.8 (7.9) 55-87 10.4 (3.1) 28.1 (1.8) 25-30 24.0 (4.9) 14-34
tests, R P R and B12, were performed to exclude metabolic causes of dementia. Computed tomography (CT) of the brain showed no evidence of focal lesions, and no patient had a Hachinski ischemic score (Hachinski et al. 1975) greater than 4. This scale has been shown to differentiate reasonably well patients with AD from patients with multi-infarct states; scores greater than 4 indicate a high probability of the latter. All had neuropsychological testing and observer-administered ratings, including the Folstein Mini-Mental State (Folstein et al. 1975), which varied from 9 to 29. This cognitive screening test has a maximum score of 30. Scores below 23 generally indicate some dementia, however, higher scores do not exclude dementia. None had received psychotropic medications for at least 10 days prior to the EEG. Group 2. Twenty-three patients with a diagnosis of major depression were studied. Each depressed patient met Research Diagnostic Criteria (Spitzer et al. 1978) for major depressive disorder, primary subtype, unipolar and non-delusional, as assessed by the Schedule for Affective Disorders and Schizophrenia (Lifetime: SADS-L). In addition, a Hamilton rating of 14 or more constituted the severity criterion for entrance into the study and in all but 3 patients, the Hamilton rating was 20 or greater. The Folstein score was required to be 25 or greater. Medical, neurologic, psychiatric and laboratory evaluations were similar to those of group 1, except that not all depressed patients had CT scans or Hachinski ratings. None had a history of drug abuse, and previous suicide attempts were rare and remote in
time. None had received psychotropic medications for at least 10 days prior to the EEG, with the exception of one subject who had received 2 mg of haloperidol 7 days prior to testing. Group 3. Sixty-one healthy control subjects with no prior history of neurologic or psychiatric illness participated in the study. As was true of the other groups, none had a history of alcoholism, severe head injury, drug abuse, cerebrovascular disease or epilepsy. They underwent the same medical, neurologic, psychiatric and laboratory evaluations as those described for group 2. All were required to have a Folstein score of 28 or greater and a Hamilton rating of 6 or less. As in the other groups, any concurrent medical problems had to be considered stable and under optimal control.
(2) Electroencephalograms and spectral analysis Sixteen-channel EEGs were performed. Disk electrodes were applied to the scalp with collodion in accordance with the international 10-20 system. A longitudinal bipolar montage was used and computerized spectral analysis was done on the following 8 derivations: C3-P3, P3-O1, C4-P4, P402, F7-T3, T3-T5, F8-T4 and T4-T6. In 2 of the derivations not used for spectral analysis ( F p l - F 7 and Fp2-F8), additional electrodes placed at the outer canthus of the eyes and referred to ipsilateral ear electrodes were used to monitor slow lateral eye movements. This was done to help ensure that the E E G samples selected for power spectral analysis included only those in which the patient was fully alert. The bandpass filters were set at 1-35 Hz (2 dB down at 35 Hz, 6 dB at 60 Hz, with a roll-off of 12 d B / o c t a v e above 60 Hz). The computer sampling rate was 128 Hz. Attempts were made throughout the recording procedure by the technologist to help reduce artifacts related to eye movement, eye blink and muscle activity. During the recording, 50-100 2 sec epochs were collected (during wakefulness with eyes closed) and stored on a DEC LSI 11/23 computer. Subsequently, the EEG was reviewed visually and 32 epochs were selected for power spectral analysis. For the parasagittal derivations (C3P3, P3-O1, C4-P4 and P4-O2), these epochs were artifact free. However, in some subjects (20 of 35
485
E E G S P E C T R A L A N A L Y S I S IN D E M E N T I A A N D D E P R E S S I O N
A D patients, 18 of 23 patients With depression and 34 of 61 controls), only the parasagittal derivations could be utilized because of excessive temporal artifact, usually muscle, or less often, eye movement artifact. Fast Fourier Transformation was applied to each of the epochs selected, and the transformed data were averaged to produce the final spectral analysis of the total 64 sec sample.
(3) Data analysis The total power of the spectrum and the relative power (% of total E E G power) in the delta (1.00-3.99 Hz), theta (4.00-7.99 Hz), alpha 1 (8.00-9.99 Hz), alpha 2 (10.00-12.99 Hz), beta 1 (13.00-19.99 Hz), and beta 2 (20.00-30.00 Hz) bands were calculated. Log transformations of the relative power of the various bandwidths in each derivation were calculated as recommended (John et al. 1980; Gasser et al. 1982), using log (x/1 - x), where x is the fraction of total power for each 2 sec sample. Analysis of covariance was performed using the log transformations, comparing the demented group separately to the control and depressed groups. Data were analyzed using an A N A C O V A where group and sex of subjects were main effects and age was used as a covariate. For the parasagittal derivations we also factored into the A N A C O V A whether or not the data from the temporal derivations were used. Similar analysis compared depressed patients to controls. The mean frequency of each 2 sec epoch was calculated according to Chotas et al. (1979) as follows: 30 Hz
E mean frequency ( 1 - 3 0 Hz) =
(F(f) × f)
f=l30 Hz Y'~ F(f) f=l
where F(f) is the power spectrum. To derive the average mean frequency, the values of the mean frequency in a given subject were summed and divided by 32 (number of 2 sec epochs). In addition to an average mean frequency for the 1-30 Hz spectrum, this measure was also calculated for a 4 - 2 0 Hz spectrum, which did not include the delta and beta 2 bands and is less likely to be
affected by artifact (Chotas et al. 1979). For the 1-30 Hz spectrum, a theta-beta difference was also obtained by subtracting the sum of beta 1 and beta 2 measures from the theta measure. Between controls and demented patients, an elementary discriminant analysis based on the mean frequency of the pooled parasagittal derivations was examined by sorting cases within sex into order, i.e., a numerical sequence of increasing frequency score, and counting the number of misclassifications that would result using different mean frequencies as cutoff points. Similar discriminant analysis between depressed and demented patients employed the total of delta and theta measures in the combined parasagittal derivations. The selection of mean frequency and the construction of a delta and theta total score was based on significant univariate test results. Moreover, mean frequency and a delta and theta total are easily understood. Multivariate discfiminant function analysis was also performed (using the SPSS computer package), employing sex and the pooled parasagittal summaries of frequency, delta, theta and beta 1+2. Demented patients were compared separately to normal and depressed groups. The automatic case classification feature was intentionally biased toward the control group, asserting a Bayesian 'prior probability' of 67% to be non-demented (Klecka 1975), so that the non-demented subjects would seldom be misclassified.
Results
Table II (comparing differences in the relative powers and mean frequency between AD patients and controls) shows that the differences between these two groups were widespread, involving the parasagittal and temporal derivations. For the temporal derivations in Table II, the sample size is reduced (27 controls and 15 patients with AD), because 34 controls and 20 demented patients were judged to have excessive temporal artifact (muscle a n d / o r eye movement). Since there were no significant differences between homologous derivations within each group, we calculated a pooled score from the 4 parasagittal derivations for each bandwidth, as well as a
R.P. BRENNER ET AL.
486
TABLE II EEG spectral differences a _ AD (n = 35) vs. controls (n = 61). Derivation
Bandwidth
Mean freq.
Thetabeta
C3-P3
1-30Hz 4-20 Hz
** ** 1,
** T
** ** I"
P3-O1
1-30Hz 4-20 Hz
** ** $
** 1'
C4-P4
1-30 Hz 4-20 Hz
** ** $
** T
P4-O2
1-30Hz 4-20 Hz
** ** {
** T
F7-T3
1-30 Hz 4-20 Hz 1-30 Hz 4-20 Hz 1-30Hz 4-20 Hz 1-30Hz 4-20 Hz
** + * * + .
T3-T5 F8-T4 T4-T6
Delta
* T
Alpha 2
Beta~
Beta 2
* . T
_
** ** $
** {
** ** 1'
_
_
** ** 1,
** ],
-
** ** T
* . T
** ** $
**
+ T
** ** ]"
+ + T
** ** $
** {
-
* T + + T + T _
,1, ** $
-
{ * T
+ T
* T
-
$ ,I,
Theta
Alpha 1
1"
**
+
-
+ 1' _
_
** ** ~" * $
+ $
+ $ +
+ $
-
a Log transformation differences of relative power in the various bandwidths and mean frequency differences. In the last 4 derivations (i.e., temporal) for AD n = 15 and for controls n = 27. + P<0.05,*P<0.01,**P<0.001. - indicates P > 0.05. The direction of the arrow indicates whether the measure was increased ( T ) or decreased ( { ) in AD patients compared to controls.
mean pooled
frequency,
for
each
group.
data from the temporal
presented,
since
a
number
of
By
contrast,
d e r i v a t i o n s is n o t subjects
in
each
group had excessive temporal artifact, thus consid-
erably
reducing
powers
of the 1-30
the
sample
size.
The
Hz bandwidth
relative
in t h e p a r a s a -
gittal region (pooled data) for the three groups are plotted which
in b a r g r a p h were
form
retransformed
i n F i g . 1, u s i n g d a t a from
the
logarithmic
60-
v a l u e s u s e d in the statistical analysis. L o g a r i t h m i c
50-
of the thirty-two
transformations £
were initially performed 2 sec e p o c h s
on each
for each
subject.
Means were calculated for each subject and subse-
o I
L,0-
quently then
30"
transformation
B w ~
for each
group.
retransformed
The
into
group
means
percentages.
Since
were log
accords a greater weight to lower
values, the re-expressed mean of the log value was
20-
lower ~0IZ~ CONTROL BB DEPRESSED
DEHENTED
than
the arithmetic
mean;
thus, the
total
p e r c e n t a g e f o r e a c h g r o u p w a s l e s s t h a n 100%. Table
III
illustrates
spectral
differences
be-
O" A
0
a1
a2
Pl
~2
FREQUENCY BAND
Fig. 1. Percent of total EEG power (estimated by retransforming logarithmic values) in the various bandwidths in the control (n = 61), depressed (n = 23) and demented (n = 35) groups (pooled parasagittal derivations). Because of the retransformation o f the data (see text), the total for each group is less than 100%.
tween AD versus controls, depression
versus AD,
and depression versus controls. Compared trol subjects, demented increase alphap
in t h e t a ,
the
patients theta-beta
difference,
and
while the mean frequency, beta 1 and beta 2
were significantly decreased. tween
to con-
had a significant
the depressed
and
The
differences be-
demented
groups,
par-
EEG SPECTRAL ANALYSIS IN DEMENTIA AND DEPRESSION
ticularly in the 1-30 Hz spectrum, involved the lower end of the spectrum, with depressed subjects having significantly less delta and theta activity and an increase in alpha 2. Depressed patients differed significantly in several bandwidths when compared to controls. Like demented patients, mean frequency, beta~ and beta 2 were decreased while the theta-beta difference and alpha~ were increased. Depressed patients also had significantly less delta activity than did controls. In all group comparisons, findings were similar in the 1-30 Hz and 4-20 Hz bandwidths. The mean EEG frequency and S.D. (pooled parasagittal derivations), for the 3 groups sorted by sex, are shown in Table IV. Male-female differences were present in all groups and these differences are the subject of another study to be reported separately. We used the parasagittal mean frequency (1-30 Hz) in the discriminant analysis in comparing demented patients and controls. We found empirically (attempting to minimize misclassifications among controls), that a post-hoc cutoff frequency of 10.56 Hz in females correctly identified 15 of 23 demented females, with only 2 subjects misclassified among the 37 controls. In a similar fashion, a post-hoc cutoff of 9.99 Hz in males correctly identified 8 of 12 demented males, with 2 subjects misclassified among 24 controls. Across sexes, these represent a 66% sensitivity
487
T A B L E IV M e a n p a r a s a g i t t a l frequencies a n d S.D. sorted by sex for 3 groups. 1-30 Hz
4 - 2 0 Hz
Mean frequency
SD.
Mean frequency
S.D.
C o n t r o l s (n = 24)
11.36
1.04
11.81
0.88
D e m e n t e d (n = 12) D e p r e s s e d (n = 8)
9.76 10.47
1.57 1.02
10.78 11.03
1.02 0.93
Females C o n t r o l s (n = 37)
12.15
0.96
12.39
0.85
D e m e n t e d (n = 23) D e p r e s s e d (n = 15)
10.15 10.97
1.94 1.23
10.88 11.33
1.29 1.06
Males
(i.e., percentage of AD patients correctly identified) and a 93% specificity (i.e., percentage of controls correctly classified) (Table VA). Similar results (using a different" cutoff frequency) were obtained with the 4-20 Hz bandwidth. Multivariate discriminant function analysis (SPSS) was also performed employing sex and the parasagittal summaries of frequency, delta, theta and beta~ +2, adjusting prior probabilities to favor classification as 'normal.' This procedure yielded classification results identical to using the parasagittal mean frequency (1-30 Hz) alone, indicating that the use of mean frequency alone was as powerful as the
TABLE III E E G spectral differences a for pooled p a r a s a g i t t a l derivations. Bandwidth
Mean freq.
Thetabeta
Delta
Theta
Alpha 1
Alpha 2
Beta1
Beta 2
** ** $
** $
* **
** J,
A D (n = 35) vs. controls (n = 61) 1-30 Hz 4 - 2 0 Hz
** ** ,1,
** T
Depression (n = 23) vs. A D (n = 35) 1-30 Hz 4 - 2 0 Hz -
-
** ],
** ** I"
+ . ~
_
* $ **
-
+ + "["
Depression (n = 23) vs. controls (n = 61) 1-30 Hz 4 - 2 0 Hz
** ** ~"
** 1"
+ $
_
** , "r
_
a L o g t r a n s f o r m a t i o n differences of relative p o w e r in the v a r i o u s b a n d w i d t h s a n d m e a n f r e q u e n c y differences. P < 0.05, * P < 0.01, ** P < 0.001. - indicates P > 0.05. F o r each 2 g r o u p c o m p a r i s o n ( A D vs. controls, d e p r e s s i o n vs. A D , a n d d e p r e s s i o n vs. controls) an a r r o w is used to indicate w h e t h e r the m e a s u r e w a s increased (I") or d e c r e a s e d (J,) w h e n c o m p a r i n g the first to the s e c o n d group.
488
R.P. B R E N N E R
combination of variables indicated above. Of the 24 demented patients with Folstein scores of 22 or less, 19 were correctly classified (79% sensitivity) by the parasagittal mean frequency (1-30 Hz), while 4 of the 11 (36% sensitivity) demented patients with scores >/23 were identified (Table VB). The mean Folstein score of the 23 AD patients correctly identified by a low parasagittal mean frequency was 16.9 (range 9-26), while the mean Folstein score of the 12 AD patients misidentified was 23 (range 17-29). Because delta and theta activity differed significantly between the demented and depressed groups, while the parasagittal mean frequency did not, the former measure was devised as a simple composite for the discriminant analysis of these groups. We derived a parasagittal delta and theta percentage by retransforming the sum of the logarithmic values for these bandwidths. We found empirically that post-hoc delta and theta scores (40% in females, 43% in males), identified 17 of 35 demented patients (49% sensitivity), with no mis-
T A B L E VI C l a s s i f i c a t i o n o f A l z h e i m e r ' s disease (n = 35) vs. d e p r e s s i o n (n = 23), b y d e l t a + t h e t a score *, c o n t r o l l e d for sex.
(A) All patients with AD Group
Delta + theta
Depr AD
C l a s s i f i c a t i o n o f A l z h e i m e r ' s disease (n = 35) vs. c o n t r o l s (n = 61), b y m e a n f r e q u e n c y *, c o n t r o l l e d f o r sex.
(A) AII patients with A D Group
C AD
Low
High
Specificity
23 18
0 17
100%
Sensitivity = 1 7 / 3 5 = 49% X 2 = 15.80, P < 0.001 r 2 = 0.27
(B) AD patients divided into subgroups according to FoL~'tein score F o l s t e i n < 22 Delta + theta
F o l s t e i n >/ 23 Delta + theta
Low
High
Specifici ty
Low
High
Specificit y
Depr AD
23 10
0 14
100%
23 8
0 3
100%
Sensitivity
= 1 4 / 2 4 = 58% = 1 9 . 1 1 , P < 0.001 = 0.41
X2 r 2
TABLE V
ET AL.
= 3 / 1 1 = 27% = 6.88, P < 0.01 = 0.20
•* T h i s p e r c e n t a g e w a s d e r i v e d b y r e t r a n s f o r m i n g the s u m of the l o g a r i t h m i c values for the p a r a s a g i t t a l d e l t a a n d t h e t a b a n d w i d t h s . T h e c u t o f f score w a s 40% in females a n d 43% in males.
M e a n freq. High
Low
Specificity
57 12
4 23
93%
15 14.
Sensitivity = 2 3 / 3 5 = 66% X 2 = 38.50, P < 0.001 r 2 = 0.40
13
(B) A D patients divided into subgroups according to Folstein score
11
•
12.
F o l s t e i n < 22 M e a n freq.
F o l s t e i n /> 2 3 M e a n freq.
Low
Specificity
High
Low
Specificity
C AD
57 5
4 19
93%
57 7
4 4
93%
Sensitivity X2 r2
= 1 9 / 2 4 = 79% = 46.01, P < 0.001 = 0.54
•
•
f/
10
" "
9
High
•
8
7
= 4 / 1 1 = 36% = 8.38, P < 0.01 = 0.12
* T h e c u t o f f f r e q u e n c y d e r i v e d f r o m the p o o l e d p a r a s a g i t t a l d e r i v a t i o n s w a s 10.56 H z in f e m a l e s a n d 9.99 H z in males.
6
o
8
ib ,2 f~ 16 18 2'0 2~ 2'~ 2'e 2~ 3'o FOLSTEIN SCORE
Fig. 2. T h e P e a r s o n l i n e a r c o r r e l a t i o n b e t w e e n the p a r a s a g i t t a l m e a n f r e q u e n c y ( 1 - 3 0 Hz) a n d the F o l s t e i n score (n = 35; r = 0.65; P < 0.001) in A D p a t i e n t s .
EEG SPECTRAL ANALYSIS IN D E M E N T I A AND DEPRESSION
classifications (100% specificity) among the 23 depressed patients (Table VIA). The multivariate discriminant function analysis was not as accurate, correctly identifying 15 of 35 patients with A D with 3 misclassifications among the depressed group. Of the 24 demented patients with Folstein scores of 22 or less, 14 were correctly classified (58% sensitivity) by the total delta and theta score, while 3 of 11 (27% sensitivity) demented patients with scores >/23 were identified (Table VIB). The mean Folstein score of the 17 A D patients identified by a high delta and theta score was 16.4 (range 9-23), while the mean Folstein score of the 18 A D patients not identified was 21.4 (range 13-29). In the demented group, various measures of the pooled parasagittal data (1-30 Hz) were compared to the Folstein score. There was a significant Pearson correlation ( P < 0.001) between the Folstein score and each of the following: mean frequency ( r = 0 . 6 5 ) , theta-beta difference ( r = -0.59), theta (r = - 0 . 5 7 ) and delta ( r = - 0 . 5 1 ) activity. Fig. 2 is a scattergram of the Folstein score and the parasagittal mean frequency (1-30 Hz). For the 4 - 2 0 Hz spectrum, the parasagittal mean frequency was analyzed and this was highly correlated with the Folstein score ( P < 0.001, r = 0.59). By contrast, in the depressed group, severity of depression, as determined by the Hamilton rating, did not correlate with spectral parameters. Discussion
We found significant spectral differences between the 3 groups. Compared to controls, demented patients show a shift of the spectrum to slower frequencies, with significant increases in theta, the theta-beta difference and alpha1 and decreases in mean frequency, beta 1 and beta 2. Demented patients also differ from depressed patients, with the latter group having significantly less activity in the low end of the spectrum (delta and theta bandwidths) and more alpha 2 activity. Like the demented group, depressed patients have significant decreases in mean frequency, beta 1 and beta 2 and increases in the theta-beta difference and alphas, when compared to controls. The increased theta-beta difference in demented patients
489
compared to controls is due to both an increase in theta and a decrease in beta I and beta 2, while in depressed patients the difference is due to a decrease in beta 1 and beta2. Since a test is of value to the clinician if it helps differentiate dementia not only from normal aging but also from other disorders in the elderly such as depression, we used spectral parameters derived from 4 pooled parasagittal derivations in a retrospective discriminate analysis. In discriminating between A D patients and controls, mean frequency was 66% sensitive and 93% specific. A combined delta and theta score had a sensitivity of 49% and a specificity of 100% when comparing the A D and depressed groups. For the group comparisons AD versus controls and AD versus depression, EEG measures used for discriminative analysis were more accurate in identifying AD patients who had lower (~< 22) Folstein scores, with a sensitivity of 79% and 58%, respectively. Our 8-channel EEG results in demented patients are similar to and extend the 1- and 2-channel occipital findings noted by others (Coben et al. 1983; Penttila et al. 1985; Visser et al. 1985). Spectral E E G measures which distinguish A D patients from controls depend upon the severity of the dementia (Coben et al. 1985). In mild dementia, there is an increase in theta and a decrease in beta activity (Coben et al. 1983, 1985), whereas with greater severity of dementia there are also decreases in alpha and increases in delta activity (Stigsby et al. 1981; Coben et al. 1985; Penttila et al. 1985). Our demented patients varied in severity (Folstein scores ranged from 9 to 29), but we did not find a significant decrease in alpha or an increase in delta activity in either the pooled parasagittal or temporal derivations. This is probably a reflection of the high percentage of our patients who had mild dementia. Unlike other investigators, we divided the alpha bandwidth into 2 components and did find that the slower component (8.00-9.99 Hz) was increased compared to controls. We also found a significant correlation between various spectral parameters and severity as measured by the Folstein score. Similar correlations have been noted by others (Canter et al. 1982; Duffy et al. 1984a; Penttila et al. 1985).
490 Like O'Conner et al. (1979) We found demented patients to have more power at frequencies below 10 Hz and depressed patients more power above 10 Hz compared to each other. Visser et al. (1985) studied elderly subjects with non-organic behavioral disorders, primarily depression, and found that their spectral findings did not differ from normals, both of which differed from the demented group. The reason for the differences in our findings and those of Visser et al. (1985) is unclear. Perhaps in some of our depressed patients, depression was part of an underlying dementia, however, this seems unlikely in view of the clinical evaluation and a mean Folstein score of 28.1. Also, our depressed group did not have a significant increase in theta activity which Coben et al. (1985) found in all stages of AD. The education levels in our depressed patients, and to a lesser extent in the demented group as well, differed from controls. This difference could be a small source of variance, perhaps related to health care or reflecting pre-morbid biologic differences among the group. Further studies are needed to confirm that elderly depressed patients have spectral differences in waking EEGs compared to controls; significant E E G sleep differences have been described (Reynolds et al. 1985). Although spectral measures show large differences between groups, there is still considerable overlap, making the classification of an individual patient difficult. Visser et al. (1985), using a P3-O1 derivation, found that a mean frequency < 6.4 Hz (1-30 Hz) correctly identified 21 of 42 A D patients with one misclassification among 28 controls, and that relative power delta > 45% identified 24 demented patients with 3 misclassifications. The prominent amount of delta activity in their demented group suggests that their patients were more severely affected than ours. Prichep et al. (1983), using 19-channel recordings and cluster analysis, correctly identified 48% of 64 m o d e r a t e / severe senile demented patients of various etiologies; there were no misclassifications among 57 controls. The same cluster correctly identified 20% of another sample of patients with mild cognitive impairment. Like Prichep et al. (1983), we included A D patients with mild cognitive impairment (Folstein scores >/23) since it is in this
R.P. BRENNER ET AL. group that the clinical diagnosis of A D is more difficult. Like those investigators, we also found spectral measures to be less sensitive in this subgroup of A D patients. Utilizing 20-channel recordings of E E G and EP data in a topographic mapping method called brain electrical activity mapping (BEAM), Duffy et al. (1984a) studied patients with presenile dementia (PSD) and senile dementia (SD). They were able to discriminate 8 of 9 patients with PSD from 15 controls. In addition, 9 of 10 subjects in each group (SD and controls) were correctly classified. In the multivariate discriminant analysis of patients with PSD, regional differences in theta, beta and auditory evoked potentials were used, while in patients with SD the localized differences involved the delta and theta bandwidths. Their high correct retrospective subject classification by stepwise discriminant function analysis reflects in part the large number of measures of both EEG and EP data (obtained in a variety of behavioral states) which were surveyed to find the best predictors among a small number of cases. We used easily derived spectral parameters from 4 pooled parasagittal derivations to discriminate dementia from normal elderly subjects and depressed patients. In both group comparisons involving AD patients (i.e., AD versus controls, A D versus depression), those demented patients with lower Folstein scores were more often correctly classified than those with higher scores. However, preliminary follow-up data suggest that of those A D patients not identified by EEG, the dementia has progressed more slowly. The degree of change does not appear to be related to the initial Folstein score but rather to the parasagittal mean frequency. We are currently assessing the value of spectral E E G measures as a predictor of rate of progression.
R6sum6
Analyse spectrale EEG par ordinateur chez le sujet normal ag~, dOment ou d~pressif On a effectu6 l'analyse spectrale par ordinateur des EEG de 35 patients atteints de la maladie
EEG SPECTRAL ANALYSIS IN DEMENTIA AND DEPRESSION d ' A l z h e i m e r p u i s c o m p a r 6 les r6SUitats h c e u x d e p a t i e n t s a t t e i n t s d e d ~ p r e s s i o n p r o f o n d e (23) et c e u x d e sujets sains ~g~s t 6 m o i n s (61). P a r r a p p o r t a u x contr61es, les p a t i e n t s d 6 m e n t s o n t p r 6 s e n t 6 u n e a u g m e n t a t i o n s i g n i f i c a t i v e d a n s les b a n d e s t h & a et alpha~ ainsi q u ' u n e p l u s g r a n d e d i f f 6 r e n c e th~ta-b&a. La fr6quence moyenne parasagittale 6tait s i g n i f i c a t i v e m e n t d i m i n u ~ e ainsi q u e l'activit~ b e t a t et b e t a 2. Les p a t i e n t s d~pressifs se dist i n g u a i e n t d e p a t i e n t s d 6 m e n t s , en p a r t i c u l i e r d a n s la p a r t i e b a s s e d u spectre, en p r ~ s e n t a n t sign i f i c a t i v e m e n t m o i n s d ' a c t i v i t 6 d e l t a et t h & a . T o u t e f o i s , c o m m e p o u r le g r o u p e des d 6 m e n t s , les p a t i e n t s d~pressifs p r 6 s e n t a i e n t , p a r r a p p o r t au t 6 m o i n s , u n e d i m i n u t i o n d e leur f r 6 q u e n c e m o y e n n e p a r a s a g i t t a l e , et de leur activit6 b~ta~ et b ~ t a 2. C h e z les p a t i e n t s d 6 m e n t s , p l u s i e u r s p a r a m + t r e s du spectre (fr6quence moyenne parasagittale, activit~ d e l t a et th~ta, d i f f & e n c e th~ta-b~ta) & a i e n t 6 t r o i t e m e n t corr616s a v e c le score d e F o l s t e i n . L a d i s c r i m i n a t i o n p a r I ' E E G a ~t6 p l u s p r e c i s e p o u r i d e n t i e r les p a t i e n t s d ~ m e n t s qui p r 6 s e n t a i e n t des scores d e F o l s t e i n faibles. This work was supported in part by NIMH Grants 00295, 37869 (Dr. Reynolds) and by NIA Grant 03705 (Dr. Boiler). The assistance of Dr. Victoria Grochocinski in data management, Patricia Lordeon, R. EEG T. in performing the EEGs and Charles M. Epstein and James T. Becker for reviewing the manuscript, is gratefully acknowledged.
References Canter, N.L., Hallett, M. and Growdon, J.H. Lecithin does not affect EEG spectral analysis or P300 in Alzheimer disease. Neurology (NY), 1982, 32: 1260-1266. Chotas, H.G., Bourne, J.R. and Teschan, P.E. Heuristic techniques in the quantification of the electroencephalogram in renal failure. Comp. biomed. Res., 1979, 12: 299-312. Coben, L.A., Danziger, W. and Berg, L. Frequency analysis of the resting awake EEG in mild senile dementia of Alzheimer type. Electroenceph. clin. Neurophysiol., 1983, 55: 372-380. Coben, L.A., Danziger, W. and Storandt, M. A longitudinal EEG study of mild senile dementia of Alzheimer type: changes at 1 year and at 2.5 years. Electroenceph. clin. Neurophysiol., 1985, 61: 101-112. Duffy, F.H., Albert, M.S. and McAnulty, G. Brain electrical activity in patients with presenile and senile dementia of the Alzheimer type. Ann. Neurol., 1984a, 16: 439-448. Duffy, F.H., Albert, M.S., McAnulty, G. and Garvey, A.J.
491
Age-related differences with brain electrical activity of healthy subjects. Ann. Neurol., 1984b, 16: 430-438. Dustman, R.E., LaMarche, J.A., Cohn, N.B., Shearer, D.E. and Talone, J.M. Power spectral analysis and cortical coupling of EEG for young and old normal adults. Neurobiol. Aging, 1985, 6: 193-198. Folstein, M.F., Folstein, S.E. and McHugh, P.R 'Mini-mental state': a practical method for grading the cognitive state of patients for the clinician. J. psychiat. Res.. 1975, 12: 189-198. Gasser, T., Bficher, P. and MScks, J. Transformations towards the normal distribution of broad band spectral parameters of the EEG. Electroenceph. clin. Neurophysiol., 1982, 53: 119-124. Gerson, I.M., John, E.R., Bartlett, F. and Koenig, V. Average evoked responses (AER) in the electroencephalographic diagnosis of the normally aging brain: a practical application. Clin. Electroenceph., 1976, 7: 77-91. Goodin, D.S. Electrophysiologic evaluation of dementia. Neurol. Clin., 1985, 3: 633-647. Hachinski, V.C., Iliff, L.D., Zilhka, E., Du Boulay, G.H., McAllister, V.L., Marshall, J., Russell, R.W.R. and Symon, L. Cerebral blood flow"in dementia. Arch. Neurol. (Chic.), 1975, 32: 632-637. Hamilton, M. Development of a rating scale for prima~ depressive illness. Brit. J. soc. clin. Psychol., 1967, 6: 278-296. John, E.R., Ahn, H., Prichep, L., Trepetin, M., Brown, D. and Kaye, H. Developmental equations for the electroencephalogram. Science, 1980, 210: 1255-1258. Klecka, W.R. Discriminant analysis. In: N.H. Nie et al. (Eds.), Statistical Package for the Social Sciences. McGraw-Hill, New York, 1975: 434-467. McKhann, G., Drachman, D., Folstein, M., Katzman, R., Price, D. and Stadlan, E.M. Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA work group under the auspices of the Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology (NY), 1984, 34: 939-944. O'Conner, K.P., Shaw, J.C. and Ongley, C.O. The EEG and differential diagnosis in psychogeriatrics. Brit. J. Psychiat., 1979, 135: 156-162. Penttila, M., Partanen, J.V., Soininen, H. and Riekkinen, P.J Quantitative analysis of occipital EEG in different stages of Alzheimer's disease. Electroenceph. clin. Neurophysiol., 1985, 60: 1-6. Prichep, L., Mont, F.G., John, E.R. and Ferris, S.H. Neurometric electroencephalographic characteristics of dementia. In: B. Reisberg (Ed.), Alzheimer's Disease: the Standard Reference. Macmillan, New York, 1983: 252-257. Reynolds, C.F., Kupfer, D.J., Taska, L.S., Hoch, C.C., Spiker, D.G., Sewitch, D.E., Zimmer, B., Marin, R.S., Nelson, J.P., Martin, D. and Morycz, R. EEG sleep in elderly depressed, demented, and healthy subjects. Biol. Psychiat., 1985, 20: 431-442. Roubicek, J. The electroencephalogram in the middle-aged and the elderly. J. Amer. geriat. Soc., 1977, 25: 145-152.
492 Spitzer, R.L., Endicott, J. and Robins, E. Research diagnostic criteria: rationale and reliability. Arch gen. Psychiat., 1978, 35: 773-782. Stigsby, B., Jbhannesson, G. and Ingvar, D.H. Regional EEG analysis and regional cerebral blood flow in Alzheimer~s and Pick's diseases. Electroenceph. clin. Neurophysiol., 1981, 51: 537-547.
R.P. BRENNER ET AL. Visser, S.L., Van Tilburg, W., Hooijer, C., Jonker, C. and De Rijke, W. Visual evoked potentials (VEPs) in senile dementia (Alzheimer type) and in non-organic behavioural disorders in the elderly: comparison with EEG parameters. Electroenceph. clin. Neurophysiol., 1985, 60: 115-121.