992
BIOL PSYCHIATRY 1992;32:992-1003
Long-Term Test-Retest Reliability of Event-Related Potentials in Normals and Alcoholics Rajita Sinha, Nancy Bernardy, and O. A. Parsons
The long-term test-retest reliability of event-related potentials (ERP) measures was examined in a group of 44 controls and 71 chronic alcoholics, retested after an average of 14 months. Correlational analyses revealed moderately significant test-retest correlations for visual and auditory target NI, N2, and P3 amplitudes, with significant correlations for NI, N2 and P3 latencies. Controls and alcoholics produced similar testretest correlations for visual and auditory ERP measures. Men and women produced equally stable ERP measures over time. Overall NI and P3 amplitudes were most reliable in both groups followed by N2 amplitude, NI and N2 latency, and P3 latency. The stability of ERP measures found in this study over a 14-month period in both normals and chronic alcoholics supports the use of ERPs in the study of normal and disordered cognitive functioning.
Introduction Event-related potentials (ERPs) are being widely used in research and clinical assessment of major psychiatric disorders (Teuting et al 1983). Essentially two types of ERPs are commonly used. Short-latency ERPs or exogenous ERPs that are evoked by events extrinsic to the nervous system and their variance is accounted for by a variation of physical stimulus parameters, such as intensity, quality, modality, etc. These include the visual evoked potentials, the brainstem auditory evoked potentials, and the short-latency somatosensory evoked potentials. The reliability of these exogenous potentials have been found to be adequate and in the range of 0.75 and 0.90 (Dustman and Beck 1963; Straumanis et al 1981; Shagass et ai 1979; for a review see Teuting et al 1983). The second type are the middle and long latency ERPs or endogenous ERPs that are also triggered by external events but whose variance is primarily determined by the particular tasks and instructions assigned to the eliciting endogenous ERPs comprising of a group of components such as the Pl, N 1, P2, N2, P3, and N4. These components are thought to index various aspects of information processing such as stimulus registration, selective attention, stimulus encoding, stimulus evaluation, and categorization and response evaluation processes (for a review, see Hillyard and Kutas 1983). Thus endog-
This research was supported in part by the National Institute of Alcohol and Alcohol Abuse Grant IROIAA06135 to Drs. R. Sinha and the late Harold L, Williams, Ph,D,, co-principal investigator. Department of Psychiatry and Behavioral Science, Oklahoma Center for Alcohol and Drug-Related Studies, University of Oklahoma Health Sciences Center, Oklahoma City, Oklahoma. Address reprint requests to Dr, Rajita Sinha, Department of Psychiatry, Yale University School of Medicine, 914 ~ Howard Avenue, New Haven, CT 06519, Received February I, 1992; revised May 30, 1992. © 1992 Society of Biological Psychiatry
0006-3223/92/$05.00
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992,32:992-1003
993
enous ERPs are currently in use to assess the cognitive and functional integrity of the brain. For example, in comparison to age-matched controls, schizophrenics show smaller NI and P3 amplitudes (Pfefferbaum et al 1984; Zahn 1986); smaller P3s have been detected in depressed patients as well (Levitt et al 1973; Roth et al 1981; Pfefferbaum et al 1984); increased P3 latency has been associated with the degree of cognitive decline in dementing illness (Polich et al 1986; Pfefferbaum et al 1984), and finally decreased NI and P3 amplitudes and increased N2 and P3 latencies have been reported in detoxified chronic alcoholics (Porjesz & Begleiter 1981, 1983; Patterson et al 1987; Parsons et al 1990), all indicative of dysfunctional information-processing functions in persons with these disorders. Although changes in ERP components are used as possible biological markers in various psychiatric disorders, the long-term reliability of endogenous ERP components in these populations is still an open question. Roth et al (i975) conducted one of the first investigations on the reliability of endogenous auditory evoked potentials in 22 normal subjects tested 5 min and 7 days apart. Median measurement reliabilities within subjects for N 1 and P3 components ranged from 0.41 to 0.49 for amplitudes and 0.39 to 0.55 for latencies. Sklare and Lynn (1984) have reported reliable P3 latency correlations (ranging from 0.84 to 0.93) in a sample of 20 young adults tested across trials within one test session and across test sessions separated by 2-4 weeks. Lewis (1984) found the point-by-point correlations of the postimulus ERP waveform to be stable within subjects from session to session, whether tested hours or 2 months apart, with greatest overall stability for bimodal presentation (r = 0.70), less for visual (r - 0.58) and least for auditory records (r - 0.45). Finally, in a recent topographical investigation examining the stability of the P300 waveform in 16 normal adults with the protocol repeated 1-2 months later, Kamiski and Blair (1989) found the P300 waveform to be very stable (with r's ranging from 0.80 to 0.90) with no loss of stability from 15 min to 1 month. Although the above studies have been important for the validation of endogenous ERP components as effective and important measures in the evaluation of cognitive capabilities, several questions regarding the issue still exist. First, what is the long-term reliability of ERP measures, that is, retested after a period of I year or more; second, how does the reliability of ERP measures from a clinical sample compare with a normative sample; third, do males and females produce equally stable ERP waveforms over time, and fourth, do the NI, N2 and P3 amplitudes and latencies produce similar reliabilities within subjects over time? The present study specifically examined these four questions in a clinical sample of posttreatment chronic alcoholics, in which information-processing deficits as measured by ERP measures have been well-established (Porjesz and Begleiter 1981, 1983, 1987; Patterson et al 1987; Parsons et al 1990), but the reliability of these measures has not been tested. They were compared to a group of nonalcoholic control subjects, and retest sessions in both normais and alcoholics were conducted on an average of 14 months later.
Method
Subjects The overall group of subjects (n = 115) comprised of 71 alcoholics (44 men; 27 women) and 44 controls (20 men; 24 women). Alcoholics were recruited from community treatment
994
BIOL PSYCHIATRY 1992;32:992-1003
R.
Sinha et al
Table 1. Means and SD of Demographic Variables for Controls and Alcholics
Groups
Average frequency c of intake days/6 months
Years of education
Drug intake°
Years of alcoholismb
13.63 4- 1.87
2.48 -4- 1.43
0
0.44 __ 0.21
35.52 4- 34.87
39.18 4. 8.96
13.00 4. 1.91
4.82 4. 2.66
13.~4 4. 8.25
4.32 4. 2.11
115.22 4. 53.20
35.48 4. 11.92
13.15 4- 1.85
2.73 .4- 1.51
0
0.28 4- 0.23
7.50 4- 13.75
37.65 4. 10.39
12.86 4. 1.69
5.50 ± 2.78
10.64 4- 6.71
4.86 4- 3.27
118.57 4. 50.26
Age
Maies Controls (n = 21) Alcoholics (n = 44) Females Controls (n - 26) Alcoholics (n = 28)
Average quantity ~ of alcohol intake/occasion
37.47 -
10.07
"Subjects are asked whether they have ever taken any of a given list of drugs either for treatment of medical problems or for recreational use. The index is a summary variable of all "yes" responses from the drug list. (Drugs include tranquilizers, amphetamines, opiates, street drugs, etc,) bMale subjects have significantly longer years of alcholism than female subjects (p -- 0.01). CAveragequantity measure is in grams of absolute ethanol consumed per kg of body weight. Average alcohol quantity and frequency measures are for the 6 months prior to treatment.
programs within a 50-mile radius of Oklahoma City. All alcoholics met the criteria for alcoholism of the National Council on Alcoholism (NCA criteria 1972) and had been deto×ified for 3 to 6 weeks prior to testing. Age, education, duration of alcoholic drinking, duration of sobriety before first testing, average quantity of alcohol per occasion, and number of days drinking occurred during 6 months before treatment are presented in Table 1. Controls were recruited by word of mouth, social organizations, and newspaper ads. Only subjects who had no history of alcohol problems and who either abstained or were social drinkers (drank fewer than 5 drinks per drinking occasion) were entered into the project. As seen in Table I, controls' drinking intake per occasion (gs/kg) was about one-tenth that of the alcoholics and they drank one.fourth to one-fifth as frequently. Other exclusion criteria for both alcoholics and controls were: a history or presence of severe psychiatric illness, positive neurological history, prolonged substance abuse other than alcohol, current ingestion of psychoactive medication, positive medical history for diseases (e.g., chronic pulmonary, liver disease, or heart disease) that could affect central nervous system (CNS) functioning. Subjects who had an alcoholic mother at birth also were excluded. A retest session was conducted 12-16 months after the initial testing. Of the original sample, 58% (n -- 200) returned for the retest session. Thus, only those subjects that returned for the retest (n - 115) were included in the present analyses. We compared the group of subjects that returned for retest to the group of subjects that did not return for retest on the ERP measures from initial testing, using one-way analysis of variance (ANOVA). No differences among the two groups were obtained. Thus, we concluded that the group that returned for retest were a representative sample of alcoholics and controls. Also of the alcoholics who returned for retest, 43 ha0. remained abstinent or drank in a light controlled manner, 28 had resumed alcoholic drinking. We found the
Reliability of ERP in Normals and Alcoholics
BIOLPSYCHIATRY 1992;32:992-1003
995
test-retest reliabilities of ERP measures among alcoholics who resumed drinking and those who remained abstinent over the 14-month period to be very similar. Thus, the retest data presented in this study includes both abstinent and relapsed alcoholics at the end of a fourteen month period. Procedure Alcoholic subjects were screened at treatment centers. Controls were screened in groups at the center. Those meeting criteria for participation in the study were then scheduled for the experimental procedures. The subjects arrived around 8 A.M. in the morning and were given an Intoxilyzer test (any person with breath alcohol readings over 0.01 was rescheduled), a pure tone audiometric testing (subjects with a hearing loss over 15 dB were not accepted into the study) and were tested for color blindness (personal with red-green color blindness were not accepted, as the ERP task involved red and green targets). ERP Procedures. After the above screening, subjects were prepared for ERP recordings. ERPs were recorded from the F3, F4, Cz, Pz, and Oz scalp locations (10-20 system, Jasper 1958) and referenced to linked earlobes. The ground was on the forehead, 3 cm anterior to Fz. Grass EC2 electrode cream was used at the interface and impedance values were kept constant at or below 5 K. Two electrooculogram (EOO) electrodes were placed on the outer cantus of each eye to monitor eye movements, and trials recorded with eye movements or other movement artifacts were automatically rejected. The Grass model 78-B polygraph was calibrated with a 20 Hz, 50 p.v peak-to-peak input signal, which resulted in an output value of plus and - 2.5 volts. Analog filters of the Grass 7P511E amplifiers were set at one-half down at 0.3 Hz and 100 Hz. A/D conversion was done with 1024 bit resolution of + and - 5 volt input as the board. Data were reduced by a Technico SS-16 microprocessor and averaged ERP vectors were stored on 8-inch floppy disks. Immediately before stimulus presentation, each electrode was sampled 250 times over a 500 msec interval to provide a baseline voltage for that particular trial. During this half-second pel~od, if the sampled electroencephalogram (EEG) exceeded + or - 400 counts (80 ~v peak-to-peak), the trial was reset and the baseline sampling rest,-~rted. This was done to ensure a relatively quiescent EEG at all electrodes for 500 msec before a stimulus was presented. Immediately after a stimulus was presented, the EEG at each of the five electrode sites was sampled at 2 msec intervals for 500 msec. If during this sampling interval the EEO amplitude exceeded the + or - 400 count range at any electrode, the trial was rejected and repeated later in the experimental run. The experimenter was informed of rejections via a monitor line on the CRT. As it turned out neither the average run time nor the number of rejected trials differed between alcoholics and controls. The ERPs were recorded in an "oddball" dual modality target-selection paradigm similar to that employed by Porjesz and Begleiter (1983). Seq,Jences of randomly interspersed visual and auditory stimuli were presented with Inter Stimulus Intervals (ISis) that varied between 0.5 and 1.5 sec. The visual stimuli were red and green circles, 1 inch in diameter and equated for luminance, presented on a CRT screen 1 m in front of ~he subject's eye. The auditory stimuli were two 50-msec tone pips (1000 and 1500
996
BIOL PSYCHIATRY 1992;32:992-1003
R. Sinha et al
Hz) equated for 40 dB SL. There were four ERP runs with a minimum of 400 trials each, so that each of the four stimuli could s e r e as the "oddball" target. The latter was presented in 20% of the trials. The task for the subject was to count the number of target stimuli in the attended morality and disregard both the nontarget in that morality and all stimuli in the nonattended modality. However, in the auditory attended condition he or she was to continue to watch the visual stimuli. A detailed description of the ERP evocation procedure and recording techniques are provided in an earlier paper (Patterson et al 1987). The same ERP recording procedures were followed for the test and retest session. A peak detection algorithm was applied to the averaged trial vectors for each stimulus condition, for electrodes Oz, Pz, and Cz, to obtain automatic computation of latencies and amplitudes of the Nl, N2, and P3 ERP components. The Nl, N2, and P3 amplitudes reported here are absolute voltages measured from the ½ sec, pre-stimulus baseline. Preliminary analyses indicated that there were no significant differences in the amplitudes or latencies of these three components between the attended target trials of the two visual and auditory stimuli. Therefore, scores were combined over the two stimuli in each modality, To reduce the number of variables in the correlational analysis, only the averaged scores for the attended "target" trials of the visual and auditory modality were included in the correlational analysis. Thus, amplitudes and latencies of the N I, N2, and P3 ERP components for the attended target trials were averaged for each run. (See Figure l for a typical response under target and nontarget, attended and nonattended conditions). For the visual attend trials were scored for the Oz, Pz, and Cz locations. For the data from the auditory modality these components were averaged at the Cz scalp location, with the exception of P3, which was also scored for the Pz site.
Results Group differences between this sample of male controls and alcoholics on the initial (end of treatment) ERP measures has been previously reported (Patterson el al 1987; Parsons et al 1990). Group differences between the retested alcoholics who resumed drinking within the 14-month period versus those who stayed abstinent as well as the retested controls are discussed elsewhere (Glenn et ai, manuscript in preparation). Therefore, the present study is focused on the within subjects stability of the ERP measures in male and female controls and alcoholics over time.
Long-term Reliability An overall zero-order correlational analysis for test-retest visual and auditory ERPs was conducted separately for the entire group on all ERP measures. Because of the large number of correlations, only those correlations that were significant at the 0.01 level will be discussed and considered significant. For the entire group, N 1, N2, and P3 amplitudes for visual "attended" target trials produced significant correlations at the 0.001 level. N I amplitude r's were 0.68 at Oz, 0.63 at Cz, and 0.67 at Pz scalp locations. N2 ampdtude r's ranged from 0.48 at Oz, 0.47 at Cz and 0.56 at Pz, whereas P3 amplitude correlations were 0.67 at Cz and 0.54 at the Pz site. Test-retest correlations for the visual ERP
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1003
997
P3 REGI O H ,t
+IG
-
. . . . . . . . . .
~
4.5. UISU~L
i
t.
.
.
.
.
o
."lr .
.
.
.
.
.
.
l
'o
,
I
I
l
i
i
',
|
t
|
|
,
,
i I
......
.
| i
. ,
¢XAHNEL:
-~---TARGET -----
ATTZHDED
l
NON
TARGET
Z
Nt
J
o
÷10
t
4.S"
. . . . . . . . . . . . . . . . . . .
AUDITORY
NOT
ATTENDED
' t
"i
+
W
CHRNN[L:
-;I . . . . . .
, tL
|e-
I 1
~ ....... s
i
i
'
! ,
RAgE EUEHT FREQUENT EU EI',IT
Io I -S
-£0
J NL
:
:
:
I
I
I
2QQ
3OQ
4OO
|
1OQ
t
it
I
5QO
141 L L I SECONDS PZ VISUAL
ELECTRODE
CHANNEL ATTENDED
Figure I. ERP waveforms at Pz for a typical subject in the dual-modality "oddball" paradigm: upper waveforms are for the attended visual channel for the targets and nontargets; lower waveforms are for the concurrent (but not simultaneously presented) nonattended auditory channel. Note the P3 amplitude response in the attended visual channel and the absence of this response in the waveform from the nonattended channel. This response is reversed when the attended channel is auditory.
component latencies, although not as high as the amplitudes, were significant for NI latency at Oz (r = 0.48, p < 0.001), Pz (r = 0.45, p < 0.001), and Cz (r = 0.31, p < 0.001) locations, for N2 at Cz (r = 0.30, p < 0.01) and Pz (r = 0.30, p < 0.01), and for P3 at Cz (r = 0.27, p < 0.01) and Pz (r = 0.23, p < 0.01) sites. For the auditory "attend" target trials, the test-retest correlations for the NI, N2, and P3 amplitudes, for alcoholics and controls combined, were significant beyond the 0.001 level, with r's of 0.71 for NI amplitude at Cz, 0.43 for N2 amplitude at Cz, and correlations for the NI (r = 0.65, p < 0.001) and N2 latencies (r = 0.58, p < 0.001) at Cz were also significant. The P3 latency produced a signifi~ant correlation at the Cz site (r = 0.32, p < 0.001) but not the Pz location. Thus, overall the test-retest correlations were highest for the NI, N2, and P3 amplitude measures in both the visual and auditory target trials.
998
R. Sinha et al
BIOL PSYCHIATRY 1992;32:992-1003
Table 2. T e s t - R e t e s t Correlations for Controls and A l c h o l i c s Visual ERPs NI Location
Oz
Cz
N2 Pz
Oz
Cz
P3
Pz
Cz
Pz
3,
Controls (n = 4 0 - 4 4 ) Amplitude 0.69 b Latency 0.54 b Alcholics (n = 6 8 - 7 1 ) Amplitude 0.67 t' Latency 0.44 b
0.67 b 0.40 °
0.711' 0.73 b
0.74 b
0.65 b
0.60 b
0.69 b
0.68 b
0.33
0.55 b
0.32
0.31
0.17
0,58 b 0,26
0.64 b 0.42 b
0.34 ° 0.12
0.38 ° 0.15
0.54:' 0.30
0.68 b 0.25
0.5{P 0.34"
0.47 ° 0.54 b
---
0.53 b 0.25
0.59 b 0.20
0.39 b 0.59 b
---
0.54 b 0.38 b
0.62 b 0.23
Auditory ERPs Controls (n = 4 0 - 4 4 ) Amplitude -n
0,66 b 0,73 b
-_
Alcoholics (n = 6 8 - 7 1 ) Amplitude -Latency --
0.77 ~' 0.6(P
---
Latency
m
m
m
m
Op < 0.01,
~p < o,oom,
Comparison of Alcoholics and Controls Next we investigated whether the test-retest correlations for ERP measures among controls and alcoholics were similar (between-groups comparison of the correlation coefficients were conducted using the formula provided in Howell 1982, p. 197). Table 2 provides the visual and auditory ERP test-retest correlations for the controls and alcoholics separately. Due to the large number of between group comparisons being made, only comparisons meeting the 0.01 level of significance are discussed. Both controls and alcoholics showed high test-retest correlations for N I and P3 amplitudes at each electrode site (p < 0.001). N2 amplitude correlations were significantly higher for controls compared to alcoholics at Oz (p < 0.01). Correlations were equally significantly between both groups at the Pz and Cz location. For the visual ERP component latencies, although controls showed higher NI latency correlations they were not significantly greater than the correlations seen in the alcoholic group. Test-retest P3 latencies in both alcoholics and controls remained poorly correlated. In the auditory "attend" target trials (Table 2), NI, N2 and P3 amplitudes were highly correlated in both controls and alcoholics beyond the 0.01 level and NI and N2 latency also produced significant test-retest correlations (p < 0.001). With the exception of P3 latency at Cz in alcoholics that produced a significant correlation (p < 0.001), no other P3 latency correlation for alcoholics or controls was significant. However, controls and alcoholics were not significantly different from each other in their test-retest correlation coefficients for any auditory ERP measure. In sum, controls as a group produced significantly higher correlations (above the 0.01 level) compared to alcoholics on only I out of 28 measures (visual and auditory combined) or 4% of the comparisons. These findings indicate that alcoholics were not different from the controls in the long-term reliability of their ERP components over a 14-month period.
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1003
999
Table 3. Visual ERP Test-Retest Correlations for Male and Female Controls and Alcoholics NI Oz
Male controls (n = 16-20) Amplitude 0.63" Latency 0.59" Female controls (n - 19-24) Amplitude 0.74 b Latency 0.47 Male alcoholics (n = 34-44) Amplitude 0.7(P Latency 0.42 a Female alcoholics (n - 22-27) Amplitude 0.64 b Latency 0.44
N2
Cz
Pz
Oz
0.65 ° 0.45
0.83 b 0.39
0.68 b 0.25
P3
Cz
Pz
Cz
Pz
0.38 0.43
0.75 b
0.59 a
0.71 b
0.68 b
0.47
0.30
0.03
0.04
0.65 b 0.74 b
0.83 b 0.28
0.55" 0.63"
0.62" 0.35
0.68 b 0.47
0.6~ 0.05
0.67 b 0.26
0.TP 0.33
0.43 ° 0.08
0.38 0.14
0.73 b 0.34
0.71 b 0.09
0.8(P 0.30
0.51 a 0.17
0.60 b 0.47 a
0.28 0.21
0.35 0.20
0.40 0.19
0.63 ° 0.26
0.30 0.40
Up < 0.01. ~p < 0.001.
Reliability Comparisons Between Males and Females The correlations for male and female controls and alcoholics were compared to assess whether they produced similar test-retest correlations. As is evident from Table 3 (visual target ERPs) and Table 4 (auditory target ERPs) each subgroup produced correlations similar to their respective overall group correlations discussed earlier. Male controls and male alcoholics produced comparable test-retest correlations for visual and auditory target ERP measures. The same was true in comparing the test-retest correlations for visual and auditory target ERP measures between female controls and female alcoholics. We also compared the subgroups to test whether males and females were equally
Table 4. Auditory ERP Test-Retest Correlations for Male and Female Controls and Alcoholics N2
Cz
Cz
Cz
Pz
0.62 °
0.45 0.03
0.65" 0.14
0.52 0.53 °
0.6W 0.40
0.56" 0.28
0.44" 0.63 b
0.46" 0.27
0.64 b 0.29
0.19 --
0.65 b 0.57 a
0.57 ° 0.09
Male controls (n = 16-20) Amplitude 0.85 b Latency 0.76 b Female controls (n - 19-24) Amplitude 0.49 Latency 0.65 b Male alcoholics (n = 34 ~A) Amplitude 0.8(P Latency 0.666 Female alcoholics (n = 23-27) Amplitude 0.67 b Latency 0.46 °p < 0.01. bp < o.ool.
P3
NI
0.41
0.39 0.50 °
1000
BIOL PSYCHIATRY 1992;32:992-1003
R. Sinha et ai
reliable in each subgroup and this was found to be true for all correlations except for the visual P3 amplitude correlation at Pz. Female alcoholics produced a significantly lower correlation (p < 0.01) for visual P3 amplitude at Pz in comparison to male alcoholics.
Reliability Comparisonsfor Individual ERP Components Finally, we tested whether the NI, N2, and P3 amplitudes and latencies were equally reliable. As there was only one significant difference in correlations between controls and alcoholics, this assessment was conducted on the overall correlations from the entire groul: presented in the long-term reliability section. N I and P3 visual amplitudes produced highest test-retest correlations and these were comparable at each electrode site. Comparisons between N I and N2 amplitude and N2 and P3 amplitudes were nonsignificant. With the exception of visual N i at Pz and visual N2 at Cz all other visual amplitude correlations were significantly higher than their corresponding latency correlations (p < 0.01). Thus, for the visual "attend" target trials NI and P3 amplitudes produced highest stability over time, followed by N2 amplitude and the NI, N2, and P3 latencies. In the auditory "attend" target condition, the N 1 and P3 amplitudes were most reliable over time, followed by N I and N2 latency, N2 amplitude, and P3 latencies.
Discussion This study was designed to test the long-term test-retest reliability of ERP measures, generated in a dual modality "oddball" paradigm, in normals and chronic alcoholics, retested after an average of 14 months. To our knowledge, this is the first study to test the reliability of endogenous ERP measures in a clinical sample and to retest subjects after l year or longer. The first specific issue examined was the long-term test-retest reliability of visual and auditory target ERP measures in a large sample of controls and alcoholics. Given the large sample size for both controls and alcoholics as well as the long time period between testing sessions, our findings indicate very good reliability estimates for the NI, N2 and P3 amplitudes for both visual and auditory target conditions. Further, the correlation coefficients reported here are comparable to the test-retest reliabilities reported in previous publications (Roth et al 1975; Lewis 1984; Sklare and Lynn 1984; Karniski and Blair 1989). Although the NI, N2, and P3 latencies were not as reliable as the amplitudes they still produced significant correlations below the 0.01 level. The next issue examined was whether the controls and alcoholics produced equally reliable ERP measures over time. The two groups were compared on the correlation coefficients for each ERP measures at each site. Only 1 out of 28 or 4% of the tests conducted indicated significantly higher correlations among controls in comparison to alcoholics. This was for visual N2 amplitude at Oz. However, as male alcoholics have been shown to have decreased visual N I and P3 amplitudes and increased N2 and P3 latencies at Pz in comparison to male controls (Porjesz and Begleiter 1983, 1987; Patterson et al 1987; Parsons et al 1990), it was reassuring to find that controls and alcoholics produced similar test-retest correlations for these measures. For the auditory target condition, controls and alcoholics were not different from each other o the test-retest correlations on any ERP measure. Overall, males and females were equally reliable in the stability of their ERP measures
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1oo3
1001
over time. This finding is of interest because unlike the males, the female alcoholics in this sample did not differ from their controls on any of the ERP measures (Parsons et al 1990). One explanation for those results is that there was no control for possible effects of the menstrual cycle on the ERP measures at either initial or follow-up testing. But this explanation is less likely because the female controls showed consistent findings across sessions and their correlations were similar to that of the males. For the female alcoholics, with ~he exception of the visual P3 amplitude correlation at Pz, correlation coefficients were equally comparable to the other groups. Therefore, the lack of information-processing deficits as measured by ERP components among these female alcoholics was not due to a lack of stability of their ERP measures. The last issue we examined was whether some ERP measures were more reliable than others in this sample. Our findings indicated that NI and P3 amplitudes were most reliable in both controls and alcoholics. These measures have been associated with channel selection and stimulus registration and evaluation and have been most sensitive to information-processing deficits among clinical populations. However, P3 latency, which is often associated with stimulus evaluation/categorization time, was one of the more variable measures in this group of subjects. Although Sklare and Lynn (1984) have reported r values ranging from 0.84 to 0.93 for test-retest P3 latencies in normal subjects tested 2-4 weeks apart, greater variability of P3 latency has been reported in the literature (Roth et al 1975). This has often been associated with differences in age groups of subjects, differences in experimental paradigms, differences in recording techniques (Fein and Turetsky 1989) and variability in the P3 waveform morphology (Sklare and Lynn 1984). Further, P3 latency is known to vary as a function of attention, arousal and memory ability (Sklare and Lynn 1984; Polich 1984; Polich et al 1983). Practice effects from the first test session and the intraindividual changes that may have occurred during the long intersession interval may have led to greater variability as well. Roth et al (1975) have suggested that a task in which subjects are actively responding to target stimuli may lead to less variability in P3 latencies. As the task in this study (silent counting of the number of "oddball" stimuli presented) was not an active reaction-time task, this may have led to increased variability in P3 latency. In conclusion, our findings indicate ~at ERP measures show good test-retest reliability over long periods of time both in normal controls and in a clinical sample of chronic, recently detoxified alcoholics. With the increasing application of ERP procedures in the evaluation of disordered cognitive functioning among clinical populations, studies testing the reliability of such techniques in these populations is essential. Whether similar results will be found with other psychiatric populations remains to be seen, but it does appear important to investigate such samples for greater generalizability of findings from ERP techniques.
References Dustman RE, Beck EC (1963): Long-term stability of visually evoked potentials in man. Science 142:1480-1481. Fein G, Turetsky B (1989): P300 latency variability in normal elderly: Effects of paradigm and measurement technique., Electroencephalogr Clin Neurophysiol 72:384-394. Glenn SW, Sinha R, Parsons OA: Recovery from alcoholism: Electrophysiologicalindices. (Manuscript in preparation).
1002
BIOL PSYCHIATRY 1992;32:992-1003
R. Sinha et al
Hillyard SA, Kutas M (1983): Electrophysiology of cognitive processing. Ann Rev Psychol 34:3361. Howell DC (1982): Statistical Methods for Psychology, Boston: Duxbury Press, p 197. Jasper HH (1958): The ten-twenty electrode system of the International Federation. Electroencephalogr Clin Neurophysiolo 20:271-275. Karniski W, Blair RC (1989): Topographical and temporal stability of the P300. Electroencephalogr Clin Neurophysiol 72:373-383. Levit AL, Sutton S, Zubin J (1973): Evoked potential correlates of information processing in psychiatric patients. Psychol Med 3:487-494. Lewis GW (1984): Temporal stability of multichannel, multimodal ERP recordings, lntJ Neurosci 25:131-144. National Council on Alcoholism (1972): Criteria for the diagnosis of alcoholism. Am J Psyc,~datry 129:127-135. Parsons OA, Sinha R, Williams HL (1990): The relationships between neuropsychologieal test performance and event-related potentials in alcoholic and non-alcoholic samples. ,Alcohol Clin Exp Res 14(5):746-.755. Patterson BW, Williams HL, McLean GA, Smith LT, Schaeffer KW (1987): Alcoholism and family history of alcoholism: Effects on visual and auditory event-related potentials. Alcohol 4:265-274. Pfefferbaum A, Wenegrat BG, Ford JM, Roth WT, Kopell BS (1984): Clinical application of the P3 component of event-related potentials II. Dementia, Depression and Schizophrenia. Electroencephalogr Clin Neurophyz'iolo 59:104-124. Polich J (1984): Methodology on P300 and correlational inference. P~ychophysiology 21(4):710711. Polich J, Howard L, Start A (1983): P300 latency correlates with digit span. Psychophysiology 20(6):665-669. Polich J, Ehlers CL, Otis S, Mandell AJ, Bloom FE (1986): P300 latency reflects the degree of cognitive decline in dementing illness. Electroencephalogr Clin Neurophysiol 63:138-144. Porjesz B, Begleiter H (1981): Human evoked brain potentials and alcohol. Alcohol Clin Exp Res 5:304-317. Porjesz B, Begleiter H (1983): Brain dysfunction and alcohol. In Kissin B, Begleiter H (eds), The Biology of Alcoholism: Vol. 7. The Pathogenesis of Alcoholism: Biological Factors. New York: Plenum Press, pp 415-483. Porjesz B, Begleitcr H (1987): Evoked brain potentials and alcoholism, in Parsons OA, Butters N, Nathan PE (eds), Neuropsychology of Alcoholism: hnplications for Diagnosis and Treatment. New York: Guilford Press, pp 45-63. Roth WT, Kopell BS, Tinklenberg JR, Huntsberger GE, Kraemer HC (1975): Reliability of the contingent negative variation and the auditory evoked potential. Electroencephalogr Clin Neurophysiol 38:45-50. Roth WT, Pfefferbaum A, Horvath TB, Berger PA, Kopell BS (1981): Auditory event-related potentials in schizophrenia and depression. Psychiatry Res 4:199-212. Shagass C, Roemer RA, Straumanis JJ, Amadeo M (1979): Temporal variability of several kinds of evoked potentials in schizophrenia. Arch Gen Psychiatry 35:1341-1351. Sklare DA, Lynn GE (1984): Latency of the P3 Event-Related Potential: Normative aspects and within-subject variability. Electroencephalogr Clin Neurophysiol 59:420-424. Snyder E, Hiilyard SA (1976): Long-latency evoked potentials to irrelevant, deviant stimuli. Behav Biol 16:319-331. Straumanis JJ, Shagass C, Roemer RA (198 I): Effects of lithium, antipsychotic and antidepressant drugs on evoked potential waveship variability. Paper presented at the Society of Biological Psychiatry Annual Meeting, April, 198 I. Teuting P, Kaskey G, Buchsbaum M, Connolly J, Pert'is C, Roemer R (1983): ERPs and psycho-
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1003
1003
pathology: Biological issues. In Karrer R, Cohen J, Teuting P (eds), Brain and Information: Event Related Potentials. New York: New York Academy of Sciences. Zahn TP (1986): Psychophysiological approaches to psychopathology. In Coles MGH, Donchin E, Porges SW (eds), Psychophysiology: Systems, Processes and Applications. New York: Guilford Press, pp 508-610.
BIOL PSYCHIATRY 1992;32:992-1003
Long-Term Test-Retest Reliability of Event-Related Potentials in Normals and Alcoholics Rajita Sinha, Nancy Bernardy, and O. A. Parsons
The long-term test-retest reliability of event-related potentials (ERP) measures was examined in a group of 44 controls and 71 chronic alcoholics, retested after an average of 14 months. Correlational analyses revealed moderately significant test-retest correlations for visual and auditory target NI, N2, and P3 amplitudes, with significant correlations for NI, N2 and P3 latencies. Controls and alcoholics produced similar testretest correlations for visual and auditory ERP measures. Men and women produced equally stable ERP measures over time. Overall NI and P3 amplitudes were most reliable in both groups followed by N2 amplitude, NI and N2 latency, and P3 latency. The stability of ERP measures found in this study over a 14-month period in both normals and chronic alcoholics supports the use of ERPs in the study of normal and disordered cognitive functioning.
Introduction Event-related potentials (ERPs) are being widely used in research and clinical assessment of major psychiatric disorders (Teuting et al 1983). Essentially two types of ERPs are commonly used. Short-latency ERPs or exogenous ERPs that are evoked by events extrinsic to the nervous system and their variance is accounted for by a variation of physical stimulus parameters, such as intensity, quality, modality, etc. These include the visual evoked potentials, the brainstem auditory evoked potentials, and the short-latency somatosensory evoked potentials. The reliability of these exogenous potentials have been found to be adequate and in the range of 0.75 and 0.90 (Dustman and Beck 1963; Straumanis et al 1981; Shagass et ai 1979; for a review see Teuting et al 1983). The second type are the middle and long latency ERPs or endogenous ERPs that are also triggered by external events but whose variance is primarily determined by the particular tasks and instructions assigned to the eliciting endogenous ERPs comprising of a group of components such as the Pl, N 1, P2, N2, P3, and N4. These components are thought to index various aspects of information processing such as stimulus registration, selective attention, stimulus encoding, stimulus evaluation, and categorization and response evaluation processes (for a review, see Hillyard and Kutas 1983). Thus endog-
This research was supported in part by the National Institute of Alcohol and Alcohol Abuse Grant IROIAA06135 to Drs. R. Sinha and the late Harold L, Williams, Ph,D,, co-principal investigator. Department of Psychiatry and Behavioral Science, Oklahoma Center for Alcohol and Drug-Related Studies, University of Oklahoma Health Sciences Center, Oklahoma City, Oklahoma. Address reprint requests to Dr, Rajita Sinha, Department of Psychiatry, Yale University School of Medicine, 914 ~ Howard Avenue, New Haven, CT 06519, Received February I, 1992; revised May 30, 1992. © 1992 Society of Biological Psychiatry
0006-3223/92/$05.00
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992,32:992-1003
993
enous ERPs are currently in use to assess the cognitive and functional integrity of the brain. For example, in comparison to age-matched controls, schizophrenics show smaller NI and P3 amplitudes (Pfefferbaum et al 1984; Zahn 1986); smaller P3s have been detected in depressed patients as well (Levitt et al 1973; Roth et al 1981; Pfefferbaum et al 1984); increased P3 latency has been associated with the degree of cognitive decline in dementing illness (Polich et al 1986; Pfefferbaum et al 1984), and finally decreased NI and P3 amplitudes and increased N2 and P3 latencies have been reported in detoxified chronic alcoholics (Porjesz & Begleiter 1981, 1983; Patterson et al 1987; Parsons et al 1990), all indicative of dysfunctional information-processing functions in persons with these disorders. Although changes in ERP components are used as possible biological markers in various psychiatric disorders, the long-term reliability of endogenous ERP components in these populations is still an open question. Roth et al (i975) conducted one of the first investigations on the reliability of endogenous auditory evoked potentials in 22 normal subjects tested 5 min and 7 days apart. Median measurement reliabilities within subjects for N 1 and P3 components ranged from 0.41 to 0.49 for amplitudes and 0.39 to 0.55 for latencies. Sklare and Lynn (1984) have reported reliable P3 latency correlations (ranging from 0.84 to 0.93) in a sample of 20 young adults tested across trials within one test session and across test sessions separated by 2-4 weeks. Lewis (1984) found the point-by-point correlations of the postimulus ERP waveform to be stable within subjects from session to session, whether tested hours or 2 months apart, with greatest overall stability for bimodal presentation (r = 0.70), less for visual (r - 0.58) and least for auditory records (r - 0.45). Finally, in a recent topographical investigation examining the stability of the P300 waveform in 16 normal adults with the protocol repeated 1-2 months later, Kamiski and Blair (1989) found the P300 waveform to be very stable (with r's ranging from 0.80 to 0.90) with no loss of stability from 15 min to 1 month. Although the above studies have been important for the validation of endogenous ERP components as effective and important measures in the evaluation of cognitive capabilities, several questions regarding the issue still exist. First, what is the long-term reliability of ERP measures, that is, retested after a period of I year or more; second, how does the reliability of ERP measures from a clinical sample compare with a normative sample; third, do males and females produce equally stable ERP waveforms over time, and fourth, do the NI, N2 and P3 amplitudes and latencies produce similar reliabilities within subjects over time? The present study specifically examined these four questions in a clinical sample of posttreatment chronic alcoholics, in which information-processing deficits as measured by ERP measures have been well-established (Porjesz and Begleiter 1981, 1983, 1987; Patterson et al 1987; Parsons et al 1990), but the reliability of these measures has not been tested. They were compared to a group of nonalcoholic control subjects, and retest sessions in both normais and alcoholics were conducted on an average of 14 months later.
Method
Subjects The overall group of subjects (n = 115) comprised of 71 alcoholics (44 men; 27 women) and 44 controls (20 men; 24 women). Alcoholics were recruited from community treatment
994
BIOL PSYCHIATRY 1992;32:992-1003
R.
Sinha et al
Table 1. Means and SD of Demographic Variables for Controls and Alcholics
Groups
Average frequency c of intake days/6 months
Years of education
Drug intake°
Years of alcoholismb
13.63 4- 1.87
2.48 -4- 1.43
0
0.44 __ 0.21
35.52 4- 34.87
39.18 4. 8.96
13.00 4. 1.91
4.82 4. 2.66
13.~4 4. 8.25
4.32 4. 2.11
115.22 4. 53.20
35.48 4. 11.92
13.15 4- 1.85
2.73 .4- 1.51
0
0.28 4- 0.23
7.50 4- 13.75
37.65 4. 10.39
12.86 4. 1.69
5.50 ± 2.78
10.64 4- 6.71
4.86 4- 3.27
118.57 4. 50.26
Age
Maies Controls (n = 21) Alcoholics (n = 44) Females Controls (n - 26) Alcoholics (n = 28)
Average quantity ~ of alcohol intake/occasion
37.47 -
10.07
"Subjects are asked whether they have ever taken any of a given list of drugs either for treatment of medical problems or for recreational use. The index is a summary variable of all "yes" responses from the drug list. (Drugs include tranquilizers, amphetamines, opiates, street drugs, etc,) bMale subjects have significantly longer years of alcholism than female subjects (p -- 0.01). CAveragequantity measure is in grams of absolute ethanol consumed per kg of body weight. Average alcohol quantity and frequency measures are for the 6 months prior to treatment.
programs within a 50-mile radius of Oklahoma City. All alcoholics met the criteria for alcoholism of the National Council on Alcoholism (NCA criteria 1972) and had been deto×ified for 3 to 6 weeks prior to testing. Age, education, duration of alcoholic drinking, duration of sobriety before first testing, average quantity of alcohol per occasion, and number of days drinking occurred during 6 months before treatment are presented in Table 1. Controls were recruited by word of mouth, social organizations, and newspaper ads. Only subjects who had no history of alcohol problems and who either abstained or were social drinkers (drank fewer than 5 drinks per drinking occasion) were entered into the project. As seen in Table I, controls' drinking intake per occasion (gs/kg) was about one-tenth that of the alcoholics and they drank one.fourth to one-fifth as frequently. Other exclusion criteria for both alcoholics and controls were: a history or presence of severe psychiatric illness, positive neurological history, prolonged substance abuse other than alcohol, current ingestion of psychoactive medication, positive medical history for diseases (e.g., chronic pulmonary, liver disease, or heart disease) that could affect central nervous system (CNS) functioning. Subjects who had an alcoholic mother at birth also were excluded. A retest session was conducted 12-16 months after the initial testing. Of the original sample, 58% (n -- 200) returned for the retest session. Thus, only those subjects that returned for the retest (n - 115) were included in the present analyses. We compared the group of subjects that returned for retest to the group of subjects that did not return for retest on the ERP measures from initial testing, using one-way analysis of variance (ANOVA). No differences among the two groups were obtained. Thus, we concluded that the group that returned for retest were a representative sample of alcoholics and controls. Also of the alcoholics who returned for retest, 43 ha0. remained abstinent or drank in a light controlled manner, 28 had resumed alcoholic drinking. We found the
Reliability of ERP in Normals and Alcoholics
BIOLPSYCHIATRY 1992;32:992-1003
995
test-retest reliabilities of ERP measures among alcoholics who resumed drinking and those who remained abstinent over the 14-month period to be very similar. Thus, the retest data presented in this study includes both abstinent and relapsed alcoholics at the end of a fourteen month period. Procedure Alcoholic subjects were screened at treatment centers. Controls were screened in groups at the center. Those meeting criteria for participation in the study were then scheduled for the experimental procedures. The subjects arrived around 8 A.M. in the morning and were given an Intoxilyzer test (any person with breath alcohol readings over 0.01 was rescheduled), a pure tone audiometric testing (subjects with a hearing loss over 15 dB were not accepted into the study) and were tested for color blindness (personal with red-green color blindness were not accepted, as the ERP task involved red and green targets). ERP Procedures. After the above screening, subjects were prepared for ERP recordings. ERPs were recorded from the F3, F4, Cz, Pz, and Oz scalp locations (10-20 system, Jasper 1958) and referenced to linked earlobes. The ground was on the forehead, 3 cm anterior to Fz. Grass EC2 electrode cream was used at the interface and impedance values were kept constant at or below 5 K. Two electrooculogram (EOO) electrodes were placed on the outer cantus of each eye to monitor eye movements, and trials recorded with eye movements or other movement artifacts were automatically rejected. The Grass model 78-B polygraph was calibrated with a 20 Hz, 50 p.v peak-to-peak input signal, which resulted in an output value of plus and - 2.5 volts. Analog filters of the Grass 7P511E amplifiers were set at one-half down at 0.3 Hz and 100 Hz. A/D conversion was done with 1024 bit resolution of + and - 5 volt input as the board. Data were reduced by a Technico SS-16 microprocessor and averaged ERP vectors were stored on 8-inch floppy disks. Immediately before stimulus presentation, each electrode was sampled 250 times over a 500 msec interval to provide a baseline voltage for that particular trial. During this half-second pel~od, if the sampled electroencephalogram (EEG) exceeded + or - 400 counts (80 ~v peak-to-peak), the trial was reset and the baseline sampling rest,-~rted. This was done to ensure a relatively quiescent EEG at all electrodes for 500 msec before a stimulus was presented. Immediately after a stimulus was presented, the EEG at each of the five electrode sites was sampled at 2 msec intervals for 500 msec. If during this sampling interval the EEO amplitude exceeded the + or - 400 count range at any electrode, the trial was rejected and repeated later in the experimental run. The experimenter was informed of rejections via a monitor line on the CRT. As it turned out neither the average run time nor the number of rejected trials differed between alcoholics and controls. The ERPs were recorded in an "oddball" dual modality target-selection paradigm similar to that employed by Porjesz and Begleiter (1983). Seq,Jences of randomly interspersed visual and auditory stimuli were presented with Inter Stimulus Intervals (ISis) that varied between 0.5 and 1.5 sec. The visual stimuli were red and green circles, 1 inch in diameter and equated for luminance, presented on a CRT screen 1 m in front of ~he subject's eye. The auditory stimuli were two 50-msec tone pips (1000 and 1500
996
BIOL PSYCHIATRY 1992;32:992-1003
R. Sinha et al
Hz) equated for 40 dB SL. There were four ERP runs with a minimum of 400 trials each, so that each of the four stimuli could s e r e as the "oddball" target. The latter was presented in 20% of the trials. The task for the subject was to count the number of target stimuli in the attended morality and disregard both the nontarget in that morality and all stimuli in the nonattended modality. However, in the auditory attended condition he or she was to continue to watch the visual stimuli. A detailed description of the ERP evocation procedure and recording techniques are provided in an earlier paper (Patterson et al 1987). The same ERP recording procedures were followed for the test and retest session. A peak detection algorithm was applied to the averaged trial vectors for each stimulus condition, for electrodes Oz, Pz, and Cz, to obtain automatic computation of latencies and amplitudes of the Nl, N2, and P3 ERP components. The Nl, N2, and P3 amplitudes reported here are absolute voltages measured from the ½ sec, pre-stimulus baseline. Preliminary analyses indicated that there were no significant differences in the amplitudes or latencies of these three components between the attended target trials of the two visual and auditory stimuli. Therefore, scores were combined over the two stimuli in each modality, To reduce the number of variables in the correlational analysis, only the averaged scores for the attended "target" trials of the visual and auditory modality were included in the correlational analysis. Thus, amplitudes and latencies of the N I, N2, and P3 ERP components for the attended target trials were averaged for each run. (See Figure l for a typical response under target and nontarget, attended and nonattended conditions). For the visual attend trials were scored for the Oz, Pz, and Cz locations. For the data from the auditory modality these components were averaged at the Cz scalp location, with the exception of P3, which was also scored for the Pz site.
Results Group differences between this sample of male controls and alcoholics on the initial (end of treatment) ERP measures has been previously reported (Patterson el al 1987; Parsons et al 1990). Group differences between the retested alcoholics who resumed drinking within the 14-month period versus those who stayed abstinent as well as the retested controls are discussed elsewhere (Glenn et ai, manuscript in preparation). Therefore, the present study is focused on the within subjects stability of the ERP measures in male and female controls and alcoholics over time.
Long-term Reliability An overall zero-order correlational analysis for test-retest visual and auditory ERPs was conducted separately for the entire group on all ERP measures. Because of the large number of correlations, only those correlations that were significant at the 0.01 level will be discussed and considered significant. For the entire group, N 1, N2, and P3 amplitudes for visual "attended" target trials produced significant correlations at the 0.001 level. N I amplitude r's were 0.68 at Oz, 0.63 at Cz, and 0.67 at Pz scalp locations. N2 ampdtude r's ranged from 0.48 at Oz, 0.47 at Cz and 0.56 at Pz, whereas P3 amplitude correlations were 0.67 at Cz and 0.54 at the Pz site. Test-retest correlations for the visual ERP
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1003
997
P3 REGI O H ,t
+IG
-
. . . . . . . . . .
~
4.5. UISU~L
i
t.
.
.
.
.
o
."lr .
.
.
.
.
.
.
l
'o
,
I
I
l
i
i
',
|
t
|
|
,
,
i I
......
.
| i
. ,
¢XAHNEL:
-~---TARGET -----
ATTZHDED
l
NON
TARGET
Z
Nt
J
o
÷10
t
4.S"
. . . . . . . . . . . . . . . . . . .
AUDITORY
NOT
ATTENDED
' t
"i
+
W
CHRNN[L:
-;I . . . . . .
, tL
|e-
I 1
~ ....... s
i
i
'
! ,
RAgE EUEHT FREQUENT EU EI',IT
Io I -S
-£0
J NL
:
:
:
I
I
I
2QQ
3OQ
4OO
|
1OQ
t
it
I
5QO
141 L L I SECONDS PZ VISUAL
ELECTRODE
CHANNEL ATTENDED
Figure I. ERP waveforms at Pz for a typical subject in the dual-modality "oddball" paradigm: upper waveforms are for the attended visual channel for the targets and nontargets; lower waveforms are for the concurrent (but not simultaneously presented) nonattended auditory channel. Note the P3 amplitude response in the attended visual channel and the absence of this response in the waveform from the nonattended channel. This response is reversed when the attended channel is auditory.
component latencies, although not as high as the amplitudes, were significant for NI latency at Oz (r = 0.48, p < 0.001), Pz (r = 0.45, p < 0.001), and Cz (r = 0.31, p < 0.001) locations, for N2 at Cz (r = 0.30, p < 0.01) and Pz (r = 0.30, p < 0.01), and for P3 at Cz (r = 0.27, p < 0.01) and Pz (r = 0.23, p < 0.01) sites. For the auditory "attend" target trials, the test-retest correlations for the NI, N2, and P3 amplitudes, for alcoholics and controls combined, were significant beyond the 0.001 level, with r's of 0.71 for NI amplitude at Cz, 0.43 for N2 amplitude at Cz, and correlations for the NI (r = 0.65, p < 0.001) and N2 latencies (r = 0.58, p < 0.001) at Cz were also significant. The P3 latency produced a signifi~ant correlation at the Cz site (r = 0.32, p < 0.001) but not the Pz location. Thus, overall the test-retest correlations were highest for the NI, N2, and P3 amplitude measures in both the visual and auditory target trials.
998
R. Sinha et al
BIOL PSYCHIATRY 1992;32:992-1003
Table 2. T e s t - R e t e s t Correlations for Controls and A l c h o l i c s Visual ERPs NI Location
Oz
Cz
N2 Pz
Oz
Cz
P3
Pz
Cz
Pz
3,
Controls (n = 4 0 - 4 4 ) Amplitude 0.69 b Latency 0.54 b Alcholics (n = 6 8 - 7 1 ) Amplitude 0.67 t' Latency 0.44 b
0.67 b 0.40 °
0.711' 0.73 b
0.74 b
0.65 b
0.60 b
0.69 b
0.68 b
0.33
0.55 b
0.32
0.31
0.17
0,58 b 0,26
0.64 b 0.42 b
0.34 ° 0.12
0.38 ° 0.15
0.54:' 0.30
0.68 b 0.25
0.5{P 0.34"
0.47 ° 0.54 b
---
0.53 b 0.25
0.59 b 0.20
0.39 b 0.59 b
---
0.54 b 0.38 b
0.62 b 0.23
Auditory ERPs Controls (n = 4 0 - 4 4 ) Amplitude -n
0,66 b 0,73 b
-_
Alcoholics (n = 6 8 - 7 1 ) Amplitude -Latency --
0.77 ~' 0.6(P
---
Latency
m
m
m
m
Op < 0.01,
~p < o,oom,
Comparison of Alcoholics and Controls Next we investigated whether the test-retest correlations for ERP measures among controls and alcoholics were similar (between-groups comparison of the correlation coefficients were conducted using the formula provided in Howell 1982, p. 197). Table 2 provides the visual and auditory ERP test-retest correlations for the controls and alcoholics separately. Due to the large number of between group comparisons being made, only comparisons meeting the 0.01 level of significance are discussed. Both controls and alcoholics showed high test-retest correlations for N I and P3 amplitudes at each electrode site (p < 0.001). N2 amplitude correlations were significantly higher for controls compared to alcoholics at Oz (p < 0.01). Correlations were equally significantly between both groups at the Pz and Cz location. For the visual ERP component latencies, although controls showed higher NI latency correlations they were not significantly greater than the correlations seen in the alcoholic group. Test-retest P3 latencies in both alcoholics and controls remained poorly correlated. In the auditory "attend" target trials (Table 2), NI, N2 and P3 amplitudes were highly correlated in both controls and alcoholics beyond the 0.01 level and NI and N2 latency also produced significant test-retest correlations (p < 0.001). With the exception of P3 latency at Cz in alcoholics that produced a significant correlation (p < 0.001), no other P3 latency correlation for alcoholics or controls was significant. However, controls and alcoholics were not significantly different from each other in their test-retest correlation coefficients for any auditory ERP measure. In sum, controls as a group produced significantly higher correlations (above the 0.01 level) compared to alcoholics on only I out of 28 measures (visual and auditory combined) or 4% of the comparisons. These findings indicate that alcoholics were not different from the controls in the long-term reliability of their ERP components over a 14-month period.
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1003
999
Table 3. Visual ERP Test-Retest Correlations for Male and Female Controls and Alcoholics NI Oz
Male controls (n = 16-20) Amplitude 0.63" Latency 0.59" Female controls (n - 19-24) Amplitude 0.74 b Latency 0.47 Male alcoholics (n = 34-44) Amplitude 0.7(P Latency 0.42 a Female alcoholics (n - 22-27) Amplitude 0.64 b Latency 0.44
N2
Cz
Pz
Oz
0.65 ° 0.45
0.83 b 0.39
0.68 b 0.25
P3
Cz
Pz
Cz
Pz
0.38 0.43
0.75 b
0.59 a
0.71 b
0.68 b
0.47
0.30
0.03
0.04
0.65 b 0.74 b
0.83 b 0.28
0.55" 0.63"
0.62" 0.35
0.68 b 0.47
0.6~ 0.05
0.67 b 0.26
0.TP 0.33
0.43 ° 0.08
0.38 0.14
0.73 b 0.34
0.71 b 0.09
0.8(P 0.30
0.51 a 0.17
0.60 b 0.47 a
0.28 0.21
0.35 0.20
0.40 0.19
0.63 ° 0.26
0.30 0.40
Up < 0.01. ~p < 0.001.
Reliability Comparisons Between Males and Females The correlations for male and female controls and alcoholics were compared to assess whether they produced similar test-retest correlations. As is evident from Table 3 (visual target ERPs) and Table 4 (auditory target ERPs) each subgroup produced correlations similar to their respective overall group correlations discussed earlier. Male controls and male alcoholics produced comparable test-retest correlations for visual and auditory target ERP measures. The same was true in comparing the test-retest correlations for visual and auditory target ERP measures between female controls and female alcoholics. We also compared the subgroups to test whether males and females were equally
Table 4. Auditory ERP Test-Retest Correlations for Male and Female Controls and Alcoholics N2
Cz
Cz
Cz
Pz
0.62 °
0.45 0.03
0.65" 0.14
0.52 0.53 °
0.6W 0.40
0.56" 0.28
0.44" 0.63 b
0.46" 0.27
0.64 b 0.29
0.19 --
0.65 b 0.57 a
0.57 ° 0.09
Male controls (n = 16-20) Amplitude 0.85 b Latency 0.76 b Female controls (n - 19-24) Amplitude 0.49 Latency 0.65 b Male alcoholics (n = 34 ~A) Amplitude 0.8(P Latency 0.666 Female alcoholics (n = 23-27) Amplitude 0.67 b Latency 0.46 °p < 0.01. bp < o.ool.
P3
NI
0.41
0.39 0.50 °
1000
BIOL PSYCHIATRY 1992;32:992-1003
R. Sinha et ai
reliable in each subgroup and this was found to be true for all correlations except for the visual P3 amplitude correlation at Pz. Female alcoholics produced a significantly lower correlation (p < 0.01) for visual P3 amplitude at Pz in comparison to male alcoholics.
Reliability Comparisonsfor Individual ERP Components Finally, we tested whether the NI, N2, and P3 amplitudes and latencies were equally reliable. As there was only one significant difference in correlations between controls and alcoholics, this assessment was conducted on the overall correlations from the entire groul: presented in the long-term reliability section. N I and P3 visual amplitudes produced highest test-retest correlations and these were comparable at each electrode site. Comparisons between N I and N2 amplitude and N2 and P3 amplitudes were nonsignificant. With the exception of visual N i at Pz and visual N2 at Cz all other visual amplitude correlations were significantly higher than their corresponding latency correlations (p < 0.01). Thus, for the visual "attend" target trials NI and P3 amplitudes produced highest stability over time, followed by N2 amplitude and the NI, N2, and P3 latencies. In the auditory "attend" target condition, the N 1 and P3 amplitudes were most reliable over time, followed by N I and N2 latency, N2 amplitude, and P3 latencies.
Discussion This study was designed to test the long-term test-retest reliability of ERP measures, generated in a dual modality "oddball" paradigm, in normals and chronic alcoholics, retested after an average of 14 months. To our knowledge, this is the first study to test the reliability of endogenous ERP measures in a clinical sample and to retest subjects after l year or longer. The first specific issue examined was the long-term test-retest reliability of visual and auditory target ERP measures in a large sample of controls and alcoholics. Given the large sample size for both controls and alcoholics as well as the long time period between testing sessions, our findings indicate very good reliability estimates for the NI, N2 and P3 amplitudes for both visual and auditory target conditions. Further, the correlation coefficients reported here are comparable to the test-retest reliabilities reported in previous publications (Roth et al 1975; Lewis 1984; Sklare and Lynn 1984; Karniski and Blair 1989). Although the NI, N2, and P3 latencies were not as reliable as the amplitudes they still produced significant correlations below the 0.01 level. The next issue examined was whether the controls and alcoholics produced equally reliable ERP measures over time. The two groups were compared on the correlation coefficients for each ERP measures at each site. Only 1 out of 28 or 4% of the tests conducted indicated significantly higher correlations among controls in comparison to alcoholics. This was for visual N2 amplitude at Oz. However, as male alcoholics have been shown to have decreased visual N I and P3 amplitudes and increased N2 and P3 latencies at Pz in comparison to male controls (Porjesz and Begleiter 1983, 1987; Patterson et al 1987; Parsons et al 1990), it was reassuring to find that controls and alcoholics produced similar test-retest correlations for these measures. For the auditory target condition, controls and alcoholics were not different from each other o the test-retest correlations on any ERP measure. Overall, males and females were equally reliable in the stability of their ERP measures
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1oo3
1001
over time. This finding is of interest because unlike the males, the female alcoholics in this sample did not differ from their controls on any of the ERP measures (Parsons et al 1990). One explanation for those results is that there was no control for possible effects of the menstrual cycle on the ERP measures at either initial or follow-up testing. But this explanation is less likely because the female controls showed consistent findings across sessions and their correlations were similar to that of the males. For the female alcoholics, with ~he exception of the visual P3 amplitude correlation at Pz, correlation coefficients were equally comparable to the other groups. Therefore, the lack of information-processing deficits as measured by ERP components among these female alcoholics was not due to a lack of stability of their ERP measures. The last issue we examined was whether some ERP measures were more reliable than others in this sample. Our findings indicated that NI and P3 amplitudes were most reliable in both controls and alcoholics. These measures have been associated with channel selection and stimulus registration and evaluation and have been most sensitive to information-processing deficits among clinical populations. However, P3 latency, which is often associated with stimulus evaluation/categorization time, was one of the more variable measures in this group of subjects. Although Sklare and Lynn (1984) have reported r values ranging from 0.84 to 0.93 for test-retest P3 latencies in normal subjects tested 2-4 weeks apart, greater variability of P3 latency has been reported in the literature (Roth et al 1975). This has often been associated with differences in age groups of subjects, differences in experimental paradigms, differences in recording techniques (Fein and Turetsky 1989) and variability in the P3 waveform morphology (Sklare and Lynn 1984). Further, P3 latency is known to vary as a function of attention, arousal and memory ability (Sklare and Lynn 1984; Polich 1984; Polich et al 1983). Practice effects from the first test session and the intraindividual changes that may have occurred during the long intersession interval may have led to greater variability as well. Roth et al (1975) have suggested that a task in which subjects are actively responding to target stimuli may lead to less variability in P3 latencies. As the task in this study (silent counting of the number of "oddball" stimuli presented) was not an active reaction-time task, this may have led to increased variability in P3 latency. In conclusion, our findings indicate ~at ERP measures show good test-retest reliability over long periods of time both in normal controls and in a clinical sample of chronic, recently detoxified alcoholics. With the increasing application of ERP procedures in the evaluation of disordered cognitive functioning among clinical populations, studies testing the reliability of such techniques in these populations is essential. Whether similar results will be found with other psychiatric populations remains to be seen, but it does appear important to investigate such samples for greater generalizability of findings from ERP techniques.
References Dustman RE, Beck EC (1963): Long-term stability of visually evoked potentials in man. Science 142:1480-1481. Fein G, Turetsky B (1989): P300 latency variability in normal elderly: Effects of paradigm and measurement technique., Electroencephalogr Clin Neurophysiol 72:384-394. Glenn SW, Sinha R, Parsons OA: Recovery from alcoholism: Electrophysiologicalindices. (Manuscript in preparation).
1002
BIOL PSYCHIATRY 1992;32:992-1003
R. Sinha et al
Hillyard SA, Kutas M (1983): Electrophysiology of cognitive processing. Ann Rev Psychol 34:3361. Howell DC (1982): Statistical Methods for Psychology, Boston: Duxbury Press, p 197. Jasper HH (1958): The ten-twenty electrode system of the International Federation. Electroencephalogr Clin Neurophysiolo 20:271-275. Karniski W, Blair RC (1989): Topographical and temporal stability of the P300. Electroencephalogr Clin Neurophysiol 72:373-383. Levit AL, Sutton S, Zubin J (1973): Evoked potential correlates of information processing in psychiatric patients. Psychol Med 3:487-494. Lewis GW (1984): Temporal stability of multichannel, multimodal ERP recordings, lntJ Neurosci 25:131-144. National Council on Alcoholism (1972): Criteria for the diagnosis of alcoholism. Am J Psyc,~datry 129:127-135. Parsons OA, Sinha R, Williams HL (1990): The relationships between neuropsychologieal test performance and event-related potentials in alcoholic and non-alcoholic samples. ,Alcohol Clin Exp Res 14(5):746-.755. Patterson BW, Williams HL, McLean GA, Smith LT, Schaeffer KW (1987): Alcoholism and family history of alcoholism: Effects on visual and auditory event-related potentials. Alcohol 4:265-274. Pfefferbaum A, Wenegrat BG, Ford JM, Roth WT, Kopell BS (1984): Clinical application of the P3 component of event-related potentials II. Dementia, Depression and Schizophrenia. Electroencephalogr Clin Neurophyz'iolo 59:104-124. Polich J (1984): Methodology on P300 and correlational inference. P~ychophysiology 21(4):710711. Polich J, Howard L, Start A (1983): P300 latency correlates with digit span. Psychophysiology 20(6):665-669. Polich J, Ehlers CL, Otis S, Mandell AJ, Bloom FE (1986): P300 latency reflects the degree of cognitive decline in dementing illness. Electroencephalogr Clin Neurophysiol 63:138-144. Porjesz B, Begleiter H (1981): Human evoked brain potentials and alcohol. Alcohol Clin Exp Res 5:304-317. Porjesz B, Begleiter H (1983): Brain dysfunction and alcohol. In Kissin B, Begleiter H (eds), The Biology of Alcoholism: Vol. 7. The Pathogenesis of Alcoholism: Biological Factors. New York: Plenum Press, pp 415-483. Porjesz B, Begleitcr H (1987): Evoked brain potentials and alcoholism, in Parsons OA, Butters N, Nathan PE (eds), Neuropsychology of Alcoholism: hnplications for Diagnosis and Treatment. New York: Guilford Press, pp 45-63. Roth WT, Kopell BS, Tinklenberg JR, Huntsberger GE, Kraemer HC (1975): Reliability of the contingent negative variation and the auditory evoked potential. Electroencephalogr Clin Neurophysiol 38:45-50. Roth WT, Pfefferbaum A, Horvath TB, Berger PA, Kopell BS (1981): Auditory event-related potentials in schizophrenia and depression. Psychiatry Res 4:199-212. Shagass C, Roemer RA, Straumanis JJ, Amadeo M (1979): Temporal variability of several kinds of evoked potentials in schizophrenia. Arch Gen Psychiatry 35:1341-1351. Sklare DA, Lynn GE (1984): Latency of the P3 Event-Related Potential: Normative aspects and within-subject variability. Electroencephalogr Clin Neurophysiol 59:420-424. Snyder E, Hiilyard SA (1976): Long-latency evoked potentials to irrelevant, deviant stimuli. Behav Biol 16:319-331. Straumanis JJ, Shagass C, Roemer RA (198 I): Effects of lithium, antipsychotic and antidepressant drugs on evoked potential waveship variability. Paper presented at the Society of Biological Psychiatry Annual Meeting, April, 198 I. Teuting P, Kaskey G, Buchsbaum M, Connolly J, Pert'is C, Roemer R (1983): ERPs and psycho-
Reliability of ERP in Normals and Alcoholics
BIOL PSYCHIATRY 1992;32:992-1003
1003
pathology: Biological issues. In Karrer R, Cohen J, Teuting P (eds), Brain and Information: Event Related Potentials. New York: New York Academy of Sciences. Zahn TP (1986): Psychophysiological approaches to psychopathology. In Coles MGH, Donchin E, Porges SW (eds), Psychophysiology: Systems, Processes and Applications. New York: Guilford Press, pp 508-610.
