Electroencephalogram analysis of electroconvulsive therapy seizures has been utilized in an attempt to tailor treatment technique to maximize clinical efficacy while minimizing cognitive side effects. For example, Krystal et al.1utilized analysis of ictal electroencephalography (EEG) indices to individualize electrical stimulus dosing and found that their algorithm, had it been used prospectively, would have allowed more patients to receive lower stimulus doses, and presumably experience less memory dysfunction, than did routine clinical care dosing. The efficacy of such an approach depends on the EEG analytic capability of the electroconvulsive therapy (ECT) machine being used. To that end, modern ECT machines produce highly sophisticated computer analyses. For example, one commonly used machine, the Thymatron System IV (Somatics, Inc., Lake Bluff, IL), has several dozen separate "EEG indices" that are routinely printed out at the end of the EEG strip for each treatment. Most of these indices, however, are unstudied in terms of their neurophysiological significance.
Several EEG studies1—6 have indicated that there may be neurophysiological differences between the seizures induced by unilateral versus bitemporal electrode placements. For example, bitemporal placement is associated with a greater degree to which the spike and wave complexes during the seizure terminate in an abrupt, flat line, termed postictal suppression.1—6 A third electrode placement, termed bifrontal, in which the electrodes are placed on either side of the forehead above the eyes, has come into use but there are no data about any possible neurophysiological differences as reflected in the ictal EEG with this placement vis a vis the other two placements.
At this stage of the use of EEG analysis in ECT practice, there are several gaps in our knowledge. One such gap, alluded to above, is the lack of research on the neurophysiological significance of the multitude of EEG indices printed out on the EEG strip. Another gap is whether there are clinically significant neurophysiological differences based on ECT technical features, such as electrode placement and stimulus dosing. As a preliminary step in advancing our understanding of EEG analysis in ECT practice, we collected the EEG printouts at several hundred ECT treatments in our facility. In this report, we investigate any differences in the numerous indices that might exist based on electrode placement.
This study was approved by the Institutional Review Board of our facility. All patients receiving ECT were included. During each ECT session, a two-channel EEG (left and right fronto-mastoid) was obtained. At the end of treatment, when the EEG is turned off, a printout appears at the end of the EEG strip enumerating the various indices. Normally, the EEG and printout are discarded. For this study, the following data were entered onto each printout: age, gender, treatment number, motor and EEG seizure duration as assessed by the treating psychiatrist, and electrode placement (right unilateral, bitemporal, or bifrontal). No patient identifying data, such as name, date of birth, or clinic identification number, were placed on the printouts.
At our facility, each patient receives a medical evaluation prior to ECT. Treatment medications consist of intravenous glycopyrrolate as antisialagogue and to reduce bradyarrhythmias, anesthesia with thiopental, and muscular paralysis with succinylcholine. Once paralysis is complete, two stimulus electrodes are placed on the head in one of three positions: bitemporal (one in each temporal fossa); bifrontal (one on each side of the forehead over each eye); or right unilateral (one in the right temporal fossa and the other just to the right of the vertex of the head). In brief, there is a long history of research regarding the effects of electrode placement on clinical efficacy and cognitive impairment.7 Electrode placement was not random but determined on a case-by-case basis by each patient’s treating psychiatrist.
ECT treatments are delivered twice or thrice weekly until clinical remission occurs. A typical course of treatments for an acute episode of depression is 6 to 10 treatments. For patients with a history of highly medication refractory, recurrent, or chronic depression, maintenance treatments may be prescribed, in which treatments are administered once every 1 to 4 weeks in an ongoing manner to prevent relapse or recurrence.
Table 1 lists the 37 outcome measures utilized in this study. These are the EEG indices on the Thymatron System IV which are based solely on EEG analysis. There are a few indices based on a combination of EEG, electrocardiogram, and EMG activity, but those were not collected in this study. The device has a computer chip capable of analyzing various aspects of the EEG signal. In brief, before the electrical stimulus is applied to the patient’s head, a baseline EEG is recorded by the machine and the data are stored in memory. After presentation of the electrical stimulus, subsequent EEG activity is compared to the baseline. A proprietary program encoded into the computer chip "reads" the EEG activity, compared to baseline, as constituting seizure activity or no seizure activity and is followed in real-time. At the point of flattening of the ictal EEG signal, the machine "reads" the seizure as being over. When the operator of the machine determines to satisfaction that EEG ictal activity has terminated, a button is pressed on the machine. At this point, the paper EEG strip terminates, and at the end of it, a printout of the indices is available for inspection. Of note, if the computer chip cannot discern baseline EEG data, due to artifact, then many of the EEG indices are not available in the printout. In this study, we only analyzed data from EEG strips in which the baseline EEG could be discerned by the machine’s computer chip.
The reader is referred to the instruction manual accompanying the machine for further details on each of the indices.8 A brief description is provided herein. In the table, "EEG 1" refers to the left fronto-mastoid channel while "EEG 2" refers to the right fronto-mastoid channel. The first nine indices are based on channel 1, with the exception of coherence measures, which are based on both channels. All other indices in the table are based on one channel as specified or both channels in the case of coherence and asymmetry measures.
The Seizure Energy Index reflects the time integral of EEG ictal amplitude over the duration of the seizure and is reported in units of squared millivolts. Presumably, this represents the "strength" of the seizure. Postictal suppression refers to the degree to which the ictal activity ends in an abrupt, easily discerned flat line versus a slow, indecipherable endpoint and is reported in percentages (100% being "perfect" postictal suppression). Maximum sustained power reflects the segment of the ictal EEG with the highest average amplitude and is reported in squared microvolts. Time to peak power, in seconds, refers to the latency from onset of seizure to point of maximal ictal EEG amplitude. Maximum sustained coherence, in percentages, reflects the highest coherence between the two EEG signals over any 3-second segment during the seizure. Time to peak coherence, in seconds, refers to latency from onset of seizure until peak interhemispheric coherence. Early, mid-, and postictal amplitude, in microvolts, refers to the mean EEG signal amplitude during the early, middle, and postictal time periods, respectively, based on prior research.1—6
The power spectral analysis makes up the rest of the EEG indices in the table. The computer chip is capable of de-artifacting the EEG signals from each channel, breaking the signals into four frequency bands (δ = 0.7 to 3.5 Hz, θ = 3.5 to 8.0 Hz, α = 8.0 to 13.0 Hz, and β = 13.0 to 25.0 Hz), and quantifying the amount of activity in each band. In the table, the absolute power indices are in units of squared microvolts. The relative power indices refer to the percentage of activity with each band and necessarily add up to 100% for the four bands for each channel.
Asymmetry and coherence, both expressed in percentages, are based on a comparison of channels 1 and 2. Coherence refers to the degree to which the two hemispheric EEG signals discharge in unison. Asymmetry is the obverse of coherence but is calculated somewhat independently as the comparison of the power between the two hemispheres while coherence is a calculation of phase synchronization between the two channels over time (David Mirkovich, Thymatron LLC, personal communication, July 28, 2006).
Moderating and Mediating Variables
The fundamental question for this study was whether electrode placement affected any of the 37 indices. We also used nine covariates in the analysis: age, gender, treatment number, motor duration of seizure, operator-observed EEG duration of seizure, machine-derived EEG duration (theoretically, the operator and machine would yield the same value but this was not always the case), and percent electrical dose on the Thymatron System IV (this reflects the amount of electricity used to elicit the seizure, which varies from patient to patient). Additionally, the machine is capable of establishing the impedance of the electrical circuit formed by the two electrodes on the head before the seizure (static impedance) and during the seizure (dynamic impedance). These two latter variables constitute the final covariates.
Data were analyzed using analysis of covariance (ANCOVA), with electrode placement being the independent variable of interest. All models adjusted for age, gender, treatment number, motor duration, EEG duration—machine- and observation-derived—% energy, static impedance, and dynamic impedance. Since there were 37 outcome variables of interest, a Bonferroni correction was used to account for multiple comparisons. Thus, p values less than 0.00135 (0.05 ÷ 37) are considered statistically significant.
The data are presented in the table. A total of 549 EEG printouts was available (171 for bifrontal, 212 for bitemporal, and 166 for unilateral). Due to incomplete data on some variables, not all of these could be used for all analyses. As the printouts were "de-identified," meaning that patient identifying information was not recorded, we do not know precisely how many patients this dataset represents, as more than one treatment (sometimes a whole course of treatments) was used if available. We estimate that, at the least, printouts from 50 to 100 patients are represented. The mean results for each of the 37 outcome measures are presented for each electrode placement followed by the p values for the comparisons between the means. Using a p value of less than 0.00135 as significant, there are a few comparisons of significance. For postictal suppression, bitemporal placement was significantly higher than that for unilateral placement. For early ictal amplitude, bifrontal placement was significantly greater than unilateral. The same was true for absolute power in EEG channel 1 in the theta range of frequencies. In the beta range, bifrontal was greater than that for both of the other placements. Relative power in EEG channel 1 in the beta range was greater for bifrontal than for bitemporal placement. For EEG channel 2, absolute power in the theta range was greater for bifrontal than for unilateral placement, while that for the alpha and beta ranges was greater for bitemporal than for unilateral placement. Also for EEG channel 2, relative power in the alpha and beta ranges was greater for bifrontal than for bitemporal placement. Finally, coherence in the alpha range was greater for bifrontal than for bitemporal placement.
Another way of looking at the data is presented in the column labeled "order" in the table. For each EEG index, a higher value is putatively considered evidence of a "stronger" or more neurophysiologically profound seizure, which presumably may correlate with clinical outcomes. The exception to this is the asymmetry indices, for which lower values putatively correspond to stronger seizures. In the "order" column in the table, we listed the order of the three electrode placement means, regardless of significance value, from "strongest" to "weakest." For the majority of indices, unilateral was "weakest."
There are two general reasons for studying quantitative EEG indices in ECT. First, such indices might correlate with clinical efficacy and thus be used to guide ECT technique to maximize efficacy and minimize cognitive side effects.1—6 Second, such study may shed light onto neurophysiological mechanisms of ECT.1—6 Electrode placement affects clinical efficacy and cognitive side effects. Attempts have been made to correlate these aspects of ECT with neurophysiological measures, such as quantitative EEG analysis. Previous research has documented consistent differences between bitemporal and unilateral electrode placement, particularly with regards to postictal suppression and early, mid-, and postictal EEG amplitude.1—6 Some of these indices also correlate with clinical efficacy. Two unstudied areas have been the quantitative EEG profile of another commonly utilized electrode placement (bifrontal), and the profiles of the majority of the several dozen indices available on modern ECT machines. In the present study, we provided, through analysis of several hundred EEG printouts from ECT treatments, mean values for the parameters on the Thymatron System IV ECT apparatus, and compared these values across the three electrode placements. As we have no clinical or cognitive outcome data, we are not in a position to add to the knowledge base regarding the correlation of EEG indices with those parameters.
In our data set, it does not appear that there are dramatic differences among the electrode placements. Significant differences by statistical tests, where applicable, do not appear to be substantially different enough so that the treating clinician could predict from a printout alone which electrode placement was used. Ultimately, the correlation between these EEG indices and clinical outcome must be studied, and a limitation of our data lies in that we do not have outcome measures available. Nonetheless, we feel that our data do represent an initial attempt to discern whether consistent EEG differences occur based on electrode placement. Though the few statistically significant comparisons do not seem to lend themselves to a cohesive summary statement, it does seem rather consistent that the unilateral placement appears "weaker" than the other two placements. Of course, one of the limitations in attempting to draw conclusions from our data is that the full underlying significance of the numerous indices studied is unknown. It remains for future data collections to enhance this knowledge base. A further limitation is that the conclusions one might draw from our data are pertinent only to the particular EEG monitoring system contained within the Thymatron System IV apparatus. Generalizability to other ECT devices or to more sophisticated EEG devices is not established.
A further challenge to the ECT field is to determine, from a neurophysiologically well-informed position, exactly what the various indices mean. There appears at present a paucity of theory to explain the significance of the majority of the 37 indices currently available on the Thymatron System IV. It appears that technology for EEG analysis in ECT has outpaced clinical practice, and future research should address this gap.
In order to establish clinical relevance of the many indices currently available on modern ECT machines, there must be outcome research utilizing quantitative assessments of antidepressant efficacy as well as cognitive effects to see if any quantitative EEG indices are associated with outcome. In the meantime, our dataset does not reveal any consistent differences among the three commonly utilized electrode placements.