Quantitative Electroencephalographic Analysis of Delirium Tremens Development Following Alcohol-withdrawal Seizure

Seizures and delirium tremens (DTs) are recognized as severe alcohol-withdrawal symptoms. Prolonged admission and serious complications associated with alcohol-withdrawal are responsible for increased costs and use of medical and social resources. We compared differences in quantitative electroencephalography (EEG) in patients after alcohol-withdrawal seizures (AWS; n = 13), performed in the intensive care unit within 48 h of admission, and in age- and sex-matched healthy controls. We also investigated the prognostic value of quantitative EEG, for the development of alcohol DTs after AWS in a retrospective, case ‒ control study. The spectral power of each band frequency and the ratio of the theta to alpha band (TAR) in the electroencephalogram were analysed using iSyncBrain ® (iMediSync, Inc., Korea). The beta frequency and the alpha frequency band power were signicantly higher and lower, respectively, in patients than in age- and sex-matched healthy controls. In AWS patients with DTs, the relative beta3 power was lower, particularly in the left frontal area, and the TAR was signicantly higher in the central channel than in those without DTs. Quantitative EEG showed neuronal excitability and decreased cognitive activities characteristic of AWS patients associated with alcohol withdrawal state and we demonstrated that quantitative EEG also might be a helpful tool for detecting patients at high risk of developing DTs during an alcohol-dependence period. Another index used to explore the difference in EEG characteristics for predicting alcohol-withdrawal delirium is the frequency band ratio. Our results showed an increased TAR in the central area in patients who developed DTs. The reverse metric of the TAR, the alpha/theta ratio (ATR), has been used as a marker of functional connectivity and performance enhancement, and is used to discriminate individuals with probable Alzheimer’s disease from healthy older controls. Increased frontal-midline theta is related to learning in a task and during performance of an attention-demanding procedure. The theta frequency has been associated with long-range functional interactions in working memory. The TAR is known to change with age and cognitive ability, even in healthy individuals. One study reported that the relationship between cognitive ability and the TAR was age-dependent, and that cognitive performance at the CZ midline location predicted the TAR 34 . Another study showed that the TAR was increased relative to controls in older adults with amnesic mild cognitive impairment 35 . Alpha/theta neurofeedback training has also been shown to have clinical benets in the treatment of alcoholism and addiction 36,37 . An increased TAR in patients with AWS, specically in those who progressed to DTs, might indicate addiction


Introduction
Alcohol-withdrawal seizures (AWS) occur in alcohol-dependent individuals, typically 1-3 days after the last drink, peak at about 48 hours, and are markedly reduced by day 5-7 of abstinence 1 .
As a rebound phenomenon, AWS are linked to the abrupt cessation of prolonged intoxication and alcohol abuse. Alcohol acts on the brain via several mechanisms. During alcohol abstinence, NMDA receptor function is enhanced, GABAergic transmission is reduced, and the dopaminergic system is dysregulated, all of which lead to withdrawal symptoms and signs 2 . Alcohol-withdrawal symptoms include tremor, insomnia, nausea or vomiting, transient hallucinations or illusions, psychomotor agitation, anxiety, and seizures. Seizures and delirium are recognized as severe forms of alcohol-withdrawal symptoms. 100-120 bpm 7,8 . In previous reports, admission with seizures per se and older age (> 70 years) increased the risk of hallucinations and delirium as a withdrawal complication in general hospitals. The biggest hazard for delirium is an increased number of days since the last drink 9,10 . However, previous studies have mainly focused on predicting the severity of alcohol-withdrawal symptoms, particularly seizures, and differed in their methodology. Thus, there is no general consensus on which factors increase the risk of DTs.
A predictive tool for the risk of DTs development after AWS would assist clinicians in making therapeutic decisions and thereby reduce patient mortality as well as the socioeconomic burden associated with alcoholism. Quantitative electroencephalography (QEEG) allows the analysis of quantitative features of brain function via oscillatory electrophysiological rhythms, such as spectral power and coherence.
Recently, various reports have emerged regarding the role of potential biomarkers of cognitive function in dementia 11 . A previous study on delirium detection using QEEG showed signi cantly increased theta and delta, and decreased alpha activity in delirium patients 12,13 , and the related delta power in the frontalparietal electrode pair between patients with and without delirium who underwent cardiothoracic surgery 13 .
In this study, we aimed to evaluate the differences in QEEG between patients who have had AWS and healthy controls, and to investigate early changes of QEEG for the development of alcohol-related DTs after AWS.

Demographic and clinical characteristics of patients
Consecutive patients who visited the emergency department between March 2018 and December 2020 for acute seizures, deemed to be related to alcohol-withdrawal by a physician, were retrospectively included in the study.
Thirteen patients were nally enrolled in the study. The mean age of participants was 50.1 ± 10.7 years and all patients were male. DTs occurred in eight of the 13 patients (61.5%). Most of them (10 of 13 patients, 76.9%) experienced their rst-ever seizure at admission; for the others, all previous seizures occurred during the alcohol-withdrawal state. AWS occurred on average 31.5 ± 21.5 h after the last selfreported drinking by participants. The average daily amount of alcohol consumption was 160.1 ± 65.9 g/day.
The demographics and clinical characteristics of the patients with and without DTs are shown in Table 1.
There were no signi cant differences in demographic and clinical alcohol-related factors, such as age, sex, history of AWS, average amount of alcohol consumed, and time interval from the last drink to seizure between the two groups. Laboratory tests also did not show statistically signi cant differences between the groups. Comparison of spectral power pattern between alcoholwithdrawal patients and healthy controls We performed a relative and absolute sensor-level analysis between the AWS patients and the age-and sex-matched healthy adult group.
In the absolute spectral power analysis, the AWS patients showed lower alpha-2 power in the bilateral temporal and occipital areas, and higher beta-1 and beta-2 power in the central (Cz) area than the healthy controls (P < 0.05).
In the relative power analysis of the AWS patients, the relative alpha-1 power in the frontocentral area and the alpha-2 power in virtually the entire cortex, except for the parietal area, were lower in the the AWS group than in healthy control group (Figure 2A-B, P < 0.05).
In contrast, the relative delta power was higher in the frontal and temporo-parieto-occipital areas ( Figure  2C, P < 0.05), and the all range of beta power (beta-1, beta-2, and beta-3) was higher in the frontal and central areas in the the AWS group than in the normal controls ( Figure 2D-F, P < 0.05).
Comparison of spectral power pattern between the patients with and without DTs To identify the difference in the spectral power pattern and nd early predictive features of DT, we compared the QEEG performed immediately (within 48 h) after AWS between the groups with (n = 8) and without DT (n = 5).
In the absolute spectral power analysis, high frequency power, particularly beta-3 power in the left frontal (Fp1 and F3), and parietal (Pz) areas were relatively lower in patients with DTs than in those without DTs (P < 0.05) The relative beta-3 powers of the left cortical areas (F3, T3, C3, and P3) were lower in the DTs group ( Figure 3A). However, the beta-3 power of the right hemisphere and other ranges of spectral powers were not signi cantly different between the two groups (P < 0.05).
In the spectral power ratios, the central (Cz) and occipital (O1 and O2) theta to alpha ratios (TAR) of DT patients were higher than those of non-DT patients (P < 0.05) ( Figure 2B).

Discussion
In this study, we compared the relative and absolute spectral power patterns across frequency bands between AWS patients in the acute period (within 48 hours after AWS) and age-and sex-matched healthy controls. In addition, we investigated whether spectral characteristics from early standard EEG are helpful for predicting and identifying patients at risk of developing DTs. The main ndings of the current study were as follows: 1) Clinical and alcohol-related variables were comparable between AWS patients with or without DTs. 2) In QEEG analysis, AWS patients demonstrated higher relative delta, theta, and all range of beta power than did the healthy control group, whereas alpha power was lower in the AWS group. The spectral differences between the groups were predominant in the frontal area (except alpha-2 power, which was lower in almost all brain areas).
3) The absolute and relative spectral analysis showed that the high-frequency beta band power, speci cally beta-3 (20-30 Hz), was lower in patients with than in patients without DTs. These differences were recorded with a certain topographic dominance in the left frontal cortical areas. Additionally, in terms of the spectral power ratios, patients with DTs had higher TAR than those without DTs in the central area.
Our observation of increased beta power and decreased alpha power in alcohol-withdrawal patients, as compared to healthy controls, was consistent with previous ndings of EEG abnormalities in the alcohol-withdrawal state. Most previous studies investigating alcohol-dependence showed EEG abnormalities, such as generalized reduction of alpha rhythm and increased power in the delta, theta, and beta activities 14 . Compared with these previous results, our results expand the understanding of association between alcohol use and EEG characteristics, considering that alcohol-withdrawal syndromes are a part of alcohol-dependence. Rebound hyperexcitability, driven by abrupt cessation from chronic alcohol exposure, might affect neuroelectric activity in the central nervous system. Changes in the glutamate/GABA balance occurs during alcohol-withdrawal periods 15 . In addition, increased beta power might re ect increasing cortical hyperexcitability after seizure as well as an alcohol-withdrawal state, resulting from an imbalance between excitatory and inhibitory neurons. This speculation is supported by the mechanisms for the generation of beta oscillations, which involves the balance between networks of excitatory pyramidal cells (AMPAergic) and inhibitory interneurons (GABAergic) 16 . The cumulative neurophysiological effects of alcohol consumption on the brain might be suggestive of our ndings. Increased absolute power in the beta frequency range of 12.5-20 Hz has been observed in alcoholdependence, over all brain locations, but prominently in the central brain region, in a previous study 17 .
Another study showed alcohol-dependence was positively correlated with absolute high beta and gamma power  in the left fronto-central-parietal leads on EEG 18 .
Alpha power has been reported to be linked with cortical arousal level and cognitive and memory performance 19,20 . In particular, the slow alpha frequency (8-10 Hz; alpha-1) is related to attentional demands, whereas the fast alpha frequency (10-12 Hz; alpha-2) mediates semantic memory and stimulus-related aspects 21 . The frontal cortex (particularly the left inferior frontal gyrus), plays a role in supporting cognitive functions that are not only speci c to language, as it has many afferent and efferent connections to all other neocortical regions (i.e., the parietal, temporal, and occipital regions, as well as to the cingulate, limbic, and basal ganglia structures 22 . The prefrontal cortex is thus the only cortical area interacting with all four sensory modalities received from olfactory sensation 20 . Alcoholism may particularly affect frontal cognitive function. In various brain imaging studies, chronic alcohol intake results in reduction of regional cerebral blood ow impairment, affecting the function of cerebral tissue in the medial frontal region, decreasing tissue metabolic rates, and affecting neurophysiology 21,23 . Alcoholdependence was associated with reduced absolute power and lower voltage of the alpha frequency in our study, which might re ect alcohol-related attentional, stimulus-reactive, and cognitive dysfunctions, as compared to in healthy controls, which was consistent with previous results about alcohol-dependence and electric de ection in the post-seizure state 24,25 . Taken together, the differential dominance in the frontal area between the AWS patients and the healthy control group can be explained by a strong connection of alcoholism to frontal lobe pathology and the post-seizure state 23 .
Studies on QEEG analysis in DTs, particularly in the alcohol-withdrawal state, remain scarce. Our study highlighted that speci c EEG characteristics could be a signi cant predictor of DTs, based on a direct comparison of EEG data obtained in the acute period post-AWS. We found that the AWS group without DTs had higher absolute and relative power in the beta-3 range than did the AWS group with DTs. Although we did not perform cognitive assessment, cognitive deterioration associated with DTs could also be an explanation for the lower beta power in the AWS patients with DTs, given that lower beta power has negative effects on memory processing, such as episodic memory encoding and retrieval 26 . Patients with alcohol-dependence with DTs had worse intellectual functioning, which was clearly observed in terms of attention and productive visual-motor coordination as compared to such patients without DTs 27 .
Findings regarding beta activity related to cognitive activity have been less consistent. However, the beta-3 frequency, ranging from 20 to 30 Hz, has speci cally been associated with cognitive processes, such as semantic speech retrieval 28 , prosodic aspects 29 , and working memory 30 . It has been proposed that highfrequency band power, not only beta-3, but also gamma activity, is an indicator of cognitive activity and that coherent oscillations in this frequency range allow the binding of distant brain regions that are necessary to allow cognitive experience 31 . Visual or auditory stimulation and cognitive activities suppress brainwaves while increasing the power of the high-frequency beta and gamma bands 32 . Hence, the decreased high beta (speci cally beta-3) power in the DT group may re ect cognitive dysfunction that precedes the development of DTs. This suggests that beta-3 power could be used as a predictive factor for the development of DTs during alcohol withdrawal. Interestingly, the difference in beta power between patients with and without DTs was prominent in the left hemisphere. While the reasons for asymmetry of EEG spectral power remain unclear, our nding may be in line with studies indicating that the mirrored reduction of left and right asymmetry might depend on the speci c cognitive domain, such as verbal (left) and visuospatial (right) functions 25,26 . The frontal lobe, particularly the left frontal gyrus, is involved in functions such as creative thinking, planning of future actions, decision-making, artistic expression, aspects of emotional behaviour, and spatial working memory 22 . It has been reported that an increased high beta frequency band (22)(23)(24)(25)(26)(27)(28)(29)(30) induced by an epileptic drug (levetiracetam) was correlated with better neuropsychological measures in patients with epilepsy 33 . Enhancing activities of the neuronal networks in the prefrontal cortex and left hippocampus also correlates with an increased beta-3 band.
Another index used to explore the difference in EEG characteristics for predicting alcohol-withdrawal delirium is the frequency band ratio. Our results showed an increased TAR in the central area in patients who developed DTs. The reverse metric of the TAR, the alpha/theta ratio (ATR), has been used as a marker of functional connectivity and performance enhancement, and is used to discriminate individuals with probable Alzheimer's disease from healthy older controls. Increased frontal-midline theta is related to learning in a task and during performance of an attention-demanding procedure. The theta frequency has been associated with long-range functional interactions in working memory. The TAR is known to change with age and cognitive ability, even in healthy individuals. One study reported that the relationship between cognitive ability and the TAR was age-dependent, and that cognitive performance at the CZ midline location predicted the TAR 34 . Another study showed that the TAR was increased relative to controls in older adults with amnesic mild cognitive impairment 35 . Alpha/theta neurofeedback training has also been shown to have clinical bene ts in the treatment of alcoholism and addiction 36,37 . An increased TAR in patients with AWS, speci cally in those who progressed to DTs, might indicate addiction and preceding cognitive impairment, which was previously found to be associated with Alzheimer's disease 29 .
The current study had several limitations. First, the study sample was small, and there might have been a selection bias, since this study was performed at a single center. Unexpectedly, most of the subjects who were initially enrolled were subsequently excluded because of structural brain lesions on neuroimaging, to avoid interference on EEG. Most of our subjects had brain lesions in cerebral cortex because alcohol intoxication is one of the strongest predictors of traumatic brain injury and a greater vascular burden 38 . Second, all subjects were male, implying that it may be di cult to generalize this nding to females. Third, we did not perform a neuropsychiatric test to identify delirium. Further studies using neuropsychiatric scores and QEEG ndings (for example, cognitive deviation to the left and right cortical function or changes in degree between groups). However, this study had several strengths. This study was performed according to a strict protocol, under well-controlled circumstances and time-lines. Furthermore, no previous report has used QEEG to predict alcohol-related DTs, particularly after AWS, which is the severe state of alcohol-withdrawal.

Conclusion
Considering the broader economic and health care burden and mortality related to DTs, early identi cation of patients at risk for developing delirium tremens is essential for early intervention, to stop the cascades of negative outcomes. QEEG might be a helpful tool for detecting early changes for development of DTs in alcohol-withdrawal patients and could help physicians to identify patients with AWS who are at high risk of developing DTs during an alcohol-dependence period.

Participants
We retrospectively reviewed the electronic medical records of all patients who were diagnosed with alcohol-withdrawal seizure at admission to two university hospitals (Hallym University Hangang and Dongtan Sacred Heart Hospital) between March 2018 and December 2020. We enrolled 38 patients aged 19 years or older who underwent initial neuroimaging, such as computed tomography and/or magnetic resonance imaging, and EEG, within 48 hours after seizure, in the intensive care unit. The diagnosis of AWS and DTs was based on the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) 39 .
We excluded patients with structural brain lesions on brain imaging that could confound EEG ndings 40 and those in whom it was not possible to evaluate DTs due to underlying language disturbance, hearing di culty, or cognitive impairment. After excluding 25 patients, 13 patients were nally analysed (Fig. 1).
For the comparison of EEG between the healthy population and patients, we selected data of age-and sex-matched healthy adults from the big data collected during the development of the EEG spectral analysis software iSyncBrain® (iMediSync, Inc., Korea).

Data processing and analysis
The selected EEG data were high-pass ltered above 1 Hz, low-pass ltered below 45 Hz, and recomputed to obtain a common average reference, o ine. In addition to visual inspection, using advanced mixtureindependent component analysis (amICA), transient and stationary artifact components originating from eye movement, muscle, or heartbeat were removed. After cleansing the EEG signals, absolute and relative sensor-level analysis using a spectopo function in EEGLAB was performed for the following eight spectral bands: δ (1-4 Hz), θ (4-8 Hz), α1 (8-10 Hz), α2 (10-12 Hz), β1 ( Source reconstructions were performed using the standardized low-resolution brain electromagnetic topography (sLORETA) plugin, using a Colin 27 Head model with 68 region-of-interest segmentations based on the Desikan and Killiany atlas 43,44 .
All the preprocessing steps, de-noising using amICA, sensor-level feature extraction, source-level feature extraction, and generation of topomap images were performed on iSyncBrain ® (iMediSync, Inc., Seoul, Korea) (https://isyncbrain.com/). Colour scale bar means difference of the interested group to the other group. In Figure 2, topomap showing the differences between the patients and control group, going toward the red side indicates that the group of interest (the patient group) has a relatively lower band frequency than the other group (the control group). Conversely, based on the colour bar, going toward the blue side represents the higher band frequencies. In Figure 3, topomap showing the differences between DT (-) and DT (+) group, going toward the red side indicates that the interested group (DT(+)) has a relatively higher band frequency than the other group (DT(-)). Conversely, based on the colour bar, going toward the blue side represents the lower band frequencies. P values toward the red side indicate statistically signi cant (P< 0.05).
The analysis protocol has been documented in previous reports 11,45 .

Statistical analysis
Continuous variables with normal distributions are presented as mean ± standard deviation, while those with non-normal distributions are presented as medians (interquartile ranges).
We compared baseline demographic information and risk factor pro les of patients with DT. We used the Mann-Whitney test for comparison of continuous variables and Fisher's exact test for comparison of categorical variables. All statistical analyses in the current study were performed using SPSS (version 24.0; IBM SPSS, Chicago, IL, USA). Statistical signi cance was considered at P < 0.05.
Statistical analyses of the EEG features were performed using MATLAB (R2017b, MathWorks, Inc.). Statistical signi cance was assumed at P < 0.05, and statistical analysis was performed automatically using the iSyncBrain® program.
Declarations Figure 1 Flow chart of patient selection Figure 2 Topomap of comparison of relative spectral power bands between the patients after AWS and healthy controls Topomap showing the differences between the patients and control group, going toward the red side indicates that the group of interest (the patient group) has a relatively lower band frequency than the other group (the control group). Conversely, based on the colour bar, going toward the blue side represents the higher band frequencies. P values toward the red side indicate statistically signi cant (P< 0.05). Topomap of comparison of relative spectral power band and theta/alpha band ratio between the patients with and without delirium tremens Topomap showing the differences between DT (-) and DT (+) group, going toward the red side indicates that the interested group (DT(+)) has a relatively higher band frequency than the other group (DT(-)).
Conversely, based on the colour bar, going toward the blue side represents the lower band frequencies.
Topomap of p-value from comparison between groups toward the red side indicate statistically signi cant (p < 0.05).