EEG Theta Power Is an Early Marker of Cognitive Decline in Dementia due to Alzheimer’s Disease

Abstract

Background:

Quantitative EEG (qEEG) power could potentially be used as a diagnostic tool for Alzheimer’s disease (AD) and may further our understanding of the pathophysiology. However, the early qEEG power changes of AD are not well understood.

Objective:

To investigate the early changes in qEEG power and the possible correlation with memory function and cerebrospinal fluid biomarkers. In addition, whether qEEG power could discriminate between AD, mild cognitive impairment (MCI), and older healthy controls (HC) at the individual level.

Methods:

Standard EEGs from 138 HC, 117 MCI, and 117 AD patients were included from six Nordic memory clinics. All EEGs were recorded consecutively before the diagnosis and were not used for the consensus diagnosis. Absolute and relative power was calculated for both eyes closed and open condition.

Results:

At group level using relative power, we found significant increases globally in the theta band and decreases in high frequency power in the temporal regions for eyes closed for AD and, to a lesser extent, for MCI compared to HC. Relative theta power was significantly correlated with multiple neuropsychological measures and had the largest correlation coefficient with total tau. At the individual level, the classification rate for AD and HC was 72.9% for relative power with eyes closed.

Conclusion:

Our findings suggest that the increase in relative theta power may be the first change in patients with dementia due to AD. At the individual level, we found a moderate classification rate for AD and HC when using EEGs alone.

Keywords

Alzheimer’s disease dementia diagnostic EEG mild cognitive impairment power

INTRODUCTION

Clinical diagnostic criteria for dementia disorders are based mainly on findings from simple assessments. However, most of the clinical criteria are not very helpful in the very early phase of a dementia disorder [1]. The use of magnetic resonance imaging (MRI) techniques, positron emission tomography (PET), and examination of the cerebrospinal fluid (CSF) have proven helpful in diagnosing Alzheimer’s disease (AD) in a very early phase [2 –6], but these methods are mainly available in academic centers, and are either expensive, require special expertise, or are invasive. Electroencephalography (EEG) is non-invasive, widely available electrophysiological monitoring method to record electrical activity of the brain and has potential also as a diagnostic marker.

One approach to analyze the EEG is by using quantitative EEG (qEEG), which is the numerical analysis of the EEG data. One type of qEEG markers is power or squared amplitude of EEG rhythmic signal, which previously has been shown as a promising marker of disease state in patients with AD [7 –10]. However, some of the studies had a low number of participants (<40 AD patients) [8, 9] or used in combination with other markers pattern recognition methodology which showed that qEEG was poor at diagnosing AD, with low specificity [11]. Other studies have shown a higher classification rate comparing AD and healthy controls (HC) [7 , 12]. Furthermore, a study has combined global theta power and left temporal theta power and found a sensitivity of 87% and a specificity of 77% [13]. Most of the large clinical studies investigating power used archived data generated during the diagnostic process and only one study has prospectively recorded EEG on patients referred to a memory clinic to study power as a classifier [11]. In the current study, we wanted to test whether qEEG power recorded prospectively from multiple centers could be a potential classifier for AD.

For any potential classifier, it is essential to consider its relationship with pathophysiological changes. In AD, studies have reported an increase in the slower frequency bands (delta and theta) and a decrease in the faster frequency bands (alpha and beta) compared with HC [8 , 14– 23] when applying qEEG power. However, the earliest changes in AD have been described as increased theta with a decrease beta and at the later stages a decrease in alpha power and an increase in delta power [14, 15]. However, the literature in this field is limited. For patients with mild cognitive impairment (MCI), which is a group that has a high risk of developing AD [24], studies have found that these patients share similar qEEG characteristics as AD [16, 25]. To better understand the early changes that make it possible to use qEEG power as a classifier, the changes in AD and MCI at the time of diagnosis should be further investigated.

To obtain a better understanding of qEEG power, studies have analyzed the correlation with working memory function and CSF markers. Working memory has been associated with the theta, alpha, and beta bands [14 –16] but most studies have investigated the relation to the Folstein score [15], which is a broad score. A newer study has showed that word list recall and word list recognition were correlated to the theta, alpha, and beta2 (22– 30 Hz) bands [16], which may due to more advanced disease. With regard to CSF markers, the literature is very sparse and only few studies have found correlations between CSF biomarkers for AD and EEG measures [26 –28]. However, no studies have looked at the regions that would be expected to be associated with AD/MCI pathology (temporal and parietal regions) or determined which frequency band may be best correlated to early AD and MCI using relative power.

In the current exploratory study, the aim was to investigate 1) whether it was possible to discriminate between AD, MCI, and HC at the individual level, 2) which changes in power were observed at the time of diagnosis with the hypothesis that changes in theta would be more pronounced than both beta and alpha, and 3) whether the early qEEG power changes in temporal or parietal regions were correlated to neuropsychological test scores or CSF markers of AD.

METHODS

Participants

The data has been used in a previous publication [29]. The patients and the HC were recruited in the NORD-EEG project from six Nordic academic memory clinics, located at the university hospitals in Copenhagen and Roskilde, Denmark; Haraldsplass Deaconess Hospital in Bergen and Oslo University Hospital, Norway; Karolinska University Hospital Huddinge, Sweden; and at Landspítali University Hospital in Reykjavik, Iceland. Each center was obliged to include a minimum of 60 patients and 20 HC. All 365 patients included in the study visited the centers for their first assessment for dementia and were recruited in most clinics consecutively. We do not have a record of how many patients declined participation. The selection of patients for NORD-EEG was based on predefined exclusion criteria: significant neurological disorder with dementia other than AD, epilepsy, major psychiatric disorder, alcohol or drug abuse. In other words, the inclusion criteria were patients referred for evaluation of cognitive symptoms, with no obvious underlying conditions. The HC comprised 146 elderly persons from NORD-EEG, who were recruited from among patients’ family members, were employees at the recruiting hospitals or were recruited through advertising. All participants gave their written consent before participating in the study.

Patients with subjective cognitive decline, Lewy body dementia, vascular dementia, Parkinson’s disease dementia, frontotemporal dementia, mixed AD and vascular dementia, were excluded from the analysis in this report. Due to poor quality or lost EEGs we had to exclude eight EEGs from patients with MCI, 15 EEGs from patients with AD, and eight EEGs from HC after the preprocessing. Some of the centers participating in the study only recorded EC segments in the EEG. As a result, the number of EC and EO segments for each group is: HC (EC: 138, EO: 110); AD (EC: 117, EO: 97); MCI (EC: 117, EO: 99). Figure 1 shows a flow diagram of the included and excluded patients. Table 1 shows the characteristics of the individuals included in the analysis.

Fig.1

Flow diagram of the number of included participants in the current study and the excluded participants after preprocessing.

Table 1

Characteristics of the participants included in the study

	HC (n = 138)	AD (n = 117)	MCI (n = 117)	p-value
				AD vs. MCI vs. HC
Mean age (SD), years	66.58 (7.61)	75.49 (7.65)	70.15 (8.13)	<0.000
Sex (female), %	60.1	60.7	53.0	0.41
Education, years (SD)	13.86 (3.65)	10.07 (3.39)	11.57 (3.92)	< 0.000
MMSE, mean (SD)	28.89 (1.35)	23.52 (3.79)	27.11 (2.16)	<0.000
Word List memory, mean (SD)	20.44 (3.99)	10.93 (3.77)	14.91 (4.19)	< 0.000
Word List Recognition, mean (SD)	7.36 (1.69)	1.39 (1.41)	3.48 (2.34)	< 0.000
Word List Recall, mean (SD)	19.46 (1.41)	15.63 (2.74)	17.87 (2.01)	< 0.000
CERAD Word List, mean (SD)	47.27 (5.71)	27.94 (6.17)	36.26 (7.29)	< 0.000
Lumbal punctures performed		32	38
Amlyoid, mean (SD)		516.34 (114.52)	876.16 (354.92)	< 0.000
Total, Tau mean (SD)		532.31 (273.85)	420.98 (230.51)	0.069
Phosphorylated Tau, mean (SD)		106 (114.52)	63.16 (26.72)	0.027
Antipsychotics	1	4	1
Antidepressants	2	18	25
Tranquilizers/hypnotics	3	6	12
Anti-dementia drugs	0	18	2
Painkillers	4	4	6

HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment; SD, standard deviation; MMSE, Mini-Mental State Examination. p-values show the differences when comparing AD, MCI, and HC.

Clinical diagnostic assessment

All patients were examined according to the standards at each memory clinic [30], which were similar at all six clinics. Clinical assessment comprised: 1) a history from the patient and an informant including cognitive function and psychiatric examination; 2) a physical examination focusing on neurological and medical status; 3) blood tests to screen for disorders that could be associated with cognitive impairment; 4) neuropsychological tests covering various cognitive domains; and 5) CT or MRI of the brain to evaluate white matter changes, general atrophy and atrophy of the medial temporal lobes, which were evaluated by a radiologist. Lumbar puncture was done in a subgroup of patients to examine amyloid-β, total tau, and phosphorylated tau protein in the cerebrospinal fluid (38 patients with MCI and 32 patients with AD). Some patients were assessed with FDG-PET, or Single-photon emission computed tomography.∥The clinical diagnoses were made at consensus meetings using the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition, Text Revision and the McKhann criteria for the diagnosis of AD, with at least two experienced doctors. The diagnosis of MCI was made using Winblad criteria [31]. All diagnoses were made independently of the qEEG results; i.e., clinicians were blinded for the EEG result and thus it was not used in the consensus diagnosis. The HC were interviewed, and histories of previous and present disorders and drug use were recorded. Any individuals with a test result on a cognitive test below one standard deviation (SD), according to their age, were excluded. Further examinations were not conducted.∥

EEG recording

EEGs were recorded using NicoletOne EEG Systems (Natus^®). The IS 10– 20 system was used for electrode placement (with 19 electrodes), and the features were evaluated using the average montage (see Supplementary Figure 1 for electrode placement). Two bipolar electro-oculography channels and one electrocardiogram channel were recorded to monitor artifacts. Most EEGs were recorded with alternating eyes closed (EC) and eyes open (EO) periods of three minutes each. The participants were alerted if they became visibly drowsy, since drowsiness influences the signal.∥

Preprocessing of EEG

The data, imported to MATLAB (Mathworks, v2016a) using the EEGLAB toolbox [32], was then divided into EC and EO segments using the events placed in the EEG, which was recorded using NicoletOne EEG Systems (Natus^®). Some of the EEGs did not contain EO segments, and the EC segments were selected from the first 10 min of the recording to prevent inclusion of segments with drowsiness or sleep. The DIPFIT toolbox [33] was then used for computerized topography of the electrodes. The excessive channels (like EKG and reference electrodes) were removed and the data was bandpass filtered from 1– 70 Hz using the pop_firws function in MATLAB, with a filter order of 2, and the Kaiser window parameter beta was estimated using a maximum passband ripple of 0.001. Afterwards, the data was bandstop filtered from 45– 55 Hz using the same settings as described above. The data was subsequently downsampled to 200 Hz, if the sampling rate was above 200 Hz. Next, the data was divided into one-second epochs and the EEGs were visually inspected. Epochs with excessive noise or artifacts were removed. Channels with excessive noise, drift, or a bad connection were then interpolated using spherical interpolation. The EEG had to have ≤ three electrodes with excessive artifacts; otherwise the EEG was excluded from the analysis. Afterwards, the EEGs were re-referenced to average and independent component analysis was performed using the extended infomax algorithm [34] for each file. The number of independent components was set at 19 but was decreased if the data rank was below 19. Components containing eye blinks, eye movement, electrocardiography artifacts, or specific line noise artifacts were removed manually. Lastly, the EEGs were visually inspected again and epochs with excessive noise or artifacts were removed. The investigator who performed the preprocessing was blinded to diagnosis.∥

QEEG calculations

EEG absolute power is the square of the amplitude (microvolts) while relative power is the percentage of total power that each frequency band occupy (i.e., absolute theta power/total power*100 = relative theta power). For calculating power, we used artifact-free resting-state EEGs and calculated absolute power over time using the built-in spectopo function from the EEGLAB toolbox. Here, the window length was a one-second epoch with a downsampled sampling rate of 200 Hz. The power was calculated in each of the following frequency bands: delta (1– 3.99 Hz); theta (4– 7.99 Hz); alpha (8– 12.99 Hz); beta1 (13– 17.99 Hz); beta2 (18– 23.99 Hz); beta3 (24– 29.99 Hz); and beta (13– 29.99 Hz). The power values were converted to decibel. For relative power, we divided each of the frequency bands with the sum of the following frequency bands: delta, theta, alpha, and beta.∥The EEGs were recorded in a clinical setting and have some high-frequency noise, which may be due to muscle as previous suggested by others [35, 36]. In addition, the data was bandstop filtered between 45– 55 Hz to remove line noise, which may lead to artefacts in the surrounding frequencies [37]. Currently, the most interesting gamma frequency is 40 Hz, which has been shown to reduce amyloid-β_{1 - 40} and amyloid-β_{1 - 42} in mouse models [38]. But the gamma band for standard scalp EEG has been suggested to be largely due to muscle activity [35]. We therefore decided not to include gamma in the current analysis.∥In addition, a study has found a high specificity and sensitivity when combining global theta power and left theta coherence [13]. To compare their findings to ours, we calculated global theta power by averaging absolute theta power for F3, F4, T3, T4, T5, T6, C3, C4, P3, P4, O1, and O2 as has previously been reported [13]. Left theta coherence (T3-T5) is a measure of connectivity between T3 and T5. To calculate left theta coherence, we took the square of the cross-spectrum of the electrodes divided by the product of the power spectra of the individual electrodes.∥

Prediction

Eight different prediction analyses were considered; EC and EO for three classes (AD, MCI, and HC) and for two classes (AD, and HC), as well as for both relative and absolute power. For all prediction analysis, we excluded the two frontal electrodes (Fp1 and Fp2), due to a large amount of noise in those channels. The dataset consisted of the number of subjects by 17 channels and seven frequency bins (delta, theta, alpha, beta1, beta2, beta3, and beta) and was turned into a data matrix based on the number of subjects, which was 119, and compressed using principal component analysis, such that 99% of the data variance was kept for the subsequent classification analysis. For the classification, we used multinomial and logistic regression for the three-class classification of AD, MCI, and HC and two-class classification of AD and HC, and MCI and HC and we quantified model prediction using leave-one-out cross-validation. We report the classification accuracies averaged over the number of observations (number of subjects), left out one at a time in the leave-one-out cross-validation procedure.∥Furthermore, we also performed prediction analysis with a dataset consisting of the number of subjects by global theta power and left alpha coherence (T3-T5) between AD and HC.∥

Statistics

All statistics were performed in MATLAB (vR2016a). To compare gender, we performed chi-squared test. For age, years of education, MMSE, and the 10-word list of the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) score and sub scores (learning, recognition, and recall) we performed one-way ANOVA for three groups (AD, MCI, and HC).∥It has been demonstrated that topography as well as power levels are different between eyes open and eyes closed segments [39]. Since we were not interested in this difference in the current study, we performed statistical tests separately between EO and EC. All the EEG data was log-transformed prior to the statistical analyses. When comparing AD, MCI, and HC we performed a MANCOVA using age, gender, years of education, and current medication as covariates. To correct for multiple comparisons, we performed Bonferroni-Holm correction for 19 electrodes, seven frequency bands, and two conditions (absolute and relative power), which equals to 266 comparisons. Afterwards, we computed post-hoc t-tests, which was considered significant with a p-value <0.05. To visualize the significant results, we used the t-values from the post-hoc test for the following comparisons: AD-HC, AD-MCI, MCI-HC. All statistical results before performing Bonferroni-Holm correction can be found in Supplementary Table 1– 8.

Correlation between qEEG and neurophysiological tests and CSF markers

The power measures were only performed for relative power in EC between the sum of the left (T3+T5) and right (T4 and T6) temporal regions, and the parietal electrodes (P3, Pz, and P4). This was done since the changes in EO were not as pronounced as in EC. If the regions were not significantly different when comparing power, we did not perform partial correlations. These regions were chosen since they are the first regions to be affected in patients with AD and are in large part related to memory function. When investigating the correlation between neuropsychological measures or CSF measures (amyloid-β, total tau, and phosphorylated tau) and power scores we performed partial correlations with age, gender, years of education, and current medication as covariates. The neuropsychological measures used were MMSE and CERAD word list including the sub scores, which included learning, delayed recall and recognition. Unfortunately, CERAD scores were missing for two HC, five patients with MCI, and 14 patients with AD. The p-values were corrected for multiple comparisons using Bonferroni-Holm correction for neuropsychological and CSF markers separately.

RESULTS

Demographics

For a full description of the demographics and comparisons between groups for AD, MCI, and HC see Table 1.

Diagnostic accuracy

We calculated the chance of AD classification when comparing AD with HC, without looking at the EEG since the group sizes are unequal: 117/(138+117) = 45.88%. This showed that AD compared to HC is less likely due to unequal group sizes.

For relative power in EC, we found an accuracy of 46.2% for three groups (HC, MCI, and AD). For absolute power, the accuracy was 53.5%. Looking at AD and HC, we found an accuracy of 72.9% (specificity = 74.64%, sensitivity = 70.94%) for relative power, and 69.0% (specificity = 80.43%, sensitivity = 55.56%) for absolute power. Looking at MCI and HC, we found an accuracy of 63.1% (specificity = 70.3%, sensitivity = 54.7%) for relative power, and 60% (specificity = 78.3%, sensitivity = 38.5%) for absolute power.

In the EO condition, we found an accuracy of 48.0% for three groups (HC, MCI, and AD) for relative power, and an accuracy of 53.3% for the same comparison for absolute power. Looking at AD and HC, we found an accuracy of 72.0% (specificity = 73.64%, sensitivity = 70.10%) for relative power, and 72.5% (specificity = 77.27%, sensitivity = 67.01%) for absolute power. For MCI and HC, we found an accuracy of 60.3% (specificity = 65.5, sensitivity = 54.5) for relative power, and 62.7% (specificity = 67.3, sensitivity = 57.6) for absolute power.

For global theta power and left alpha coherence (T3-T5) between AD and HC, we found an accuracy of 54.1% (specificity = 89.86%, sensitivity = 11.97%).

Relative power for AD, MCI, and HC

In the EC condition, we found that relative power was significantly different for all electrodes (except Cz) in the theta frequency band when looking at AD versus HC, and MCI versus HC. The increase in relative theta power was significantly more pronounced in AD compared to MCI when looking at the P3, Pz, P4, F3, F4, and Fp1. For alpha power, the most pronounced decrease was found when comparing AD versus HC and the location were almost exclusively in the outermost electrodes on the head (F7, F8, T3, T4, T5, and T6). For all beta bands, we found the largest decrease in AD versus HC, with fewer differences in MCI versus HC and a few sporadic differences between AD versus MCI. However, the decrease in AD versus HC were only present in T3 in the beta3 band; see Fig. 2.

Fig.2

The relative power difference between groups (AD-HC, MCI-HC, and AD-MCI) for all frequency bands with eyes closed. The difference is illustrated if the MANCOVA was significant after performing Bonferroni-Holm correction and post-hoc t-tests. Red indicates increased relative power in the first group listed. Blue indicates decreased relative power in the first group listed. T-values were used to illustrate the difference and the color bar shows the corresponding t-value. HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

In the EO condition, we found an increase in relative theta power for AD versus HC, MCI versus HC, and AD versus MCI. The increases were mostly centered at the parietal and left temporal electrodes. In addition, we found a decrease in power in the beta1 band for AD versus HC, MCI versus HC, and AD versus MCI. The largest decreases were found for AD versus HC; see Fig. 3.

Fig.3

The relative power difference between groups (AD-HC, MCI-HC, and AD-MCI) for each frequency band with eyes open. The difference is illustrated if the MANCOVA was significant after performing Bonferroni-Holm correction and post-hoc t-tests. Red indicates increased relative power in the first group listed. Blue indicates decreased relative power in the first group listed. T-values were used to illustrate the difference and the color bar shows the corresponding t-value. HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

Absolute power for AD, MCI, and HC

In the EC condition, we found the most significant increases in the delta band for AD versus HC. Here the increases were mostly centered at the right side (P4, C4, F4). The same increases were found for AD versus MCI but to a lesser extent. No changes were found between MCI versus HC; see Fig. 4.

Fig.4

The absolute power difference between groups (AD-HC, MCI-HC, and AD-MCI) for each frequency band for both eyes closed and open. The difference is illustrated if the MANCOVA was significant after performing Bonferroni-Holm correction and post-hoc t-tests. Red indicates increased relative power in the first group listed. Blue indicates decreased relative power in the first group listed. T-values were used to illustrate the difference and the colorbar shows the corresponding t-value. HC, healthy controls; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

In the EO condition, we found only few significant electrodes (<2), which were all centered in the posterior electrodes; see Fig. 4.

Relationship between power and neuropsychological measures

Significant negative correlations were found between the theta power in both left temporal, right temporal, and parietal electrodes and MMSE, learning, recall, recognition and the total CERAD score; see Table 2. In the alpha band, we found positive correlations between left and right temporal areas and MMSE score, and between right temporal area and 10-word list recognition. For beta, beta1, and beta2, we found significant positive correlations between left temporal area and 10-word list memory, word list recognition, and CERAD word list score; see Table 2.

Table 2

Correlation coefficients between the relative power in EC condition and neuropsychological tests and CSF markers. Correlation coefficient (rho) from partial correlations

	MMSE	Word List	Word List	Word List	CERAD Word	Amyloid	Total	Phosphorylated
		Memory	Recognition	Recall	List Score		Tau	Tau
Lt T theta	-0.20^*	-0.21^*	-0.26^*	-0.07	-0.22^*	-0.08	0.24	0.02
Rt T theta	-0.17^*	-0.20^*	-0.26^*	-0.09	-0.22^*	-0.12	0.22	0.04
P theta	-0.22^*	-0.23^*	-0.27^*	-0.10	-0.24^*	-0.08	0.15	0.06
Lt T alpha	0.17^*	0.12	0.18^*	0.14	0.16^*	0.21	-0.14	-0.15
Rt T alpha	0.17^*	0.14	0.21^*	0.16^*	0.19^*	0.18	-0.16	-0.14
Lt T Beta	0.09	0.21^*	0.22^*	0.05	0.20^*	0.06	-0.12	0.04
Lt T Beta1	0.15^*	0.26^*	0.26^*	0.07	0.24^*	0.15	-0.17	0.06
Lt T Beta2	0.07	0.17^*	0.20^*	0.04	0.17^*	0.04	-0.11	0.03

^*Displays significance after Bonferroni-Holm correction for multiple comparisons.

Association between power and CSF markers

No significant correlations were found between relative power in the theta band and the CSF markers. However, the highest correlation coefficient was found between the left temporal electrodes and total tau from the CSF but was not significant; see Table 2.

DISCUSSION

In the present study, we wanted to investigate whether qEEG power could discriminate between AD, MCI, and HC at individual level. We also wanted to examine the changes in qEEG power at the time of diagnosis according to the hypothesis that changes in theta would be more pronounced than both beta and alpha in the early phase of the disease. Lastly, we wanted to investigate whether the early qEEG power changes in temporal or parietal regions were correlated to neuropsychological test scores or CSF markers. We found significant increases globally in the theta band and temporal decreases in high frequency power for eyes closed for AD and, to a lesser extent, for MCI compared to HC. Relative theta power was significantly correlated with multiple neuropsychological measures and had the largest correlation coefficient with total tau. At the individual level, the classification rate for AD and HC was 72.9% for relative power with eyes closed. The current study, the largest multicenter study to date investigating EEG power in memory clinic patients, demonstrated that theta power may be the hallmark of early clinical manifestation of neurodegeneration and that qEEG power has a moderate classification rate.

The changes in relative power for EC (see Fig. 2) were in line with results of previous research [8 , 14– 23] but less pronounced except for the theta band, where we found a global increase in relative power in AD and MCI. The increases in relative theta power could be the first sign of neurodegeneration and a sign of the underlying network dysfunction in patients with AD. In support of this, the theta band has been associated with the EEG default mode network [40]. Other studies have analyzed the association between qEEG and blood oxygen level dependence (BOLD) signal from fMRI, which demonstrated an inverse relationship between theta power and BOLD signal [41]. In addition, studies investigating rodents have found that theta waves are generated from the hippocampus [42, 43], which recently has been found in humans [44]. We therefore suggest that the increases in relative theta power should be viewed as the hallmark qEEG power sign of early AD due to the pronounced changes and link to hippocampal function. It may be corresponding to the decreases in the default mode network in AD, which previously has been shown using resting state functional MRI [45]. The further progression of the disease may lead to decreases in the beta bands and especially the lower beta1 band. The changes in absolute power in EC in the delta band may be due to the large variability and reflect the patients with more advanced AD.

Relative theta power being the first change in power in AD is also supported by significant correlations to almost all the 10-word list CERAD scores, except for the word list recall, which may be due to a floor effect. However, stronger correlation coefficients were seen in both beta and especially beta1, which may be due to beta being affected more in the later stages. Other studies have found that the most involved frequency bands in working memory includes theta, alpha, and beta [14– 16 , 46]. But this may be due to the studies investigating more advanced stages of the disease with lower MMSE scores [16] or longitudinal studies [14, 25] where the decreases in alpha and beta become more pronounced. Furthermore, the strongest correlation with CSF markers were between left temporal electrodes in the relative theta band and total tau, which supports the hypothesis that theta is the most affected frequency band at the time of diagnosis. The absence of correlations between power and amyloid-β may be due to amyloid-β reaching a plateau at this stage in the disease in patients with AD [47] with total tau still increasing.

We showed a moderate classification rate when looking at AD and HC, with a classification rate of 72.9% for relative power in EC, with specificity higher than sensitivity. The group of patients who had EO segments in the EEG unfortunately had a lower classification rate than for EC, but the classification rate between AD and HC was 72.5% for absolute power and 72.0% for relative power. These results show a better classifier than a previous study [11], and our study has a higher number of participants than some other studies [8, 9]. However, some studies have shown even higher classification rates [7 , 12], but two of them included fewer participants than in the current study [7, 10]. Another study found a promising classifier by combining left alpha coherence (T3-T5) and global theta power [13] but we found an accuracy of 54.1% when combining the two values and were not able to replicate their finding. This may be due to the study comparing AD to cognitively unimpaired depressed patients while we compared AD to HC. In the current study, the EEGs were performed independently of the diagnosis, which means they can more precisely depict the accuracy of using power to predict AD than studies using already diagnosed patients, since it has been found that power changes in patients with AD over time [14, 48]. However, most large clinical studies investigating power [7, 9] used archived data generated during the diagnostic process, but these studies did not recruit patients consecutively, as in the present study. In addition, we included all EEG features in one model, reducing the chance of false positive results when performing multiple analyses. Even though we found a moderate classification rate, power may be used to characterize patients with AD. One study showed that AD patients who were characterized as AD using cortical sources showed signs of impaired global cognition and brain structural integrity [49]. Therefore, it may be useful as a diagnostic tool to characterize the patients in the more advanced stages of AD and to provide an additional tool for monitoring progression in these patients. The classification rates are not as high as in FDG-PET, CSF biomarkers, or MRI, which has been shown to be >80% [50], but relative power may be a valuable tool in centers that do not have access to any of these methods. In addition, qEEG power may add to increase the diagnostic accuracy. However, qEEG analyses and artefact removal with independent component are largely used in research settings and their application in the clinical environment is not possible without training and education in software and procedures. Furthermore, this study used a simple linear classifier; however, we expect that more advanced classification procedures, if combined with more data, can improve the results. The ability for qEEG power to discriminate between AD, MCI, and HC showed a poor classification rate, which most likely is due to the heterogeneity of the MCI group, which showed almost equal classification between the three groups. In the current study, the MCI group was not classified as purely amnestic MCI and some of the MCI patients may already have had preclinical AD.

CSF markers were available for only 30% of the included patients, which limited the statistical power of the correlations between theta power and CSF markers. The lack of follow-up data in the MCI group prevented us from investigating pre-dementia MCI. Furthermore, the EEGs in the current study were EEGs in a clinical setting, which means that high-frequency noise was present in the gamma band and may have influenced the lower frequency bands. When looking at the demographics, we found that the HC group was younger than both AD and MCI, and the AD and MCI groups were treated with medications which may impact the EEG, especially in slower frequencies. Also, the HC group had a higher level of education than either clinical group which may also play a role in memory impairment. However, we tried to correct for these confounders by including age, medication and education as covariates when performing MANCOVA. The current study is the largest prospective multicenter study with consecutively recruited patients at an early stage of AD.

Conclusions

In the present study, we found that increased theta power was the most pronounced change in patients with dementia due to AD at the time of diagnosis and this increase may be associated with hippocampal function and the decreases found in the default mode network. Furthermore, the relative theta band was significantly correlated to working memory measures and showed the strongest correlation to total tau, which supports our initial hypothesis. In addition, the beta1 showed more pronounced changes and was better correlated to neuropsychological measures than the broad beta band, which should be considered in future studies. Lastly, we found that the best classification rate for AD and HC was for relative power with eyes closed with an accuracy of 72.9%. The qEEG power can be a helpful tool when more advanced methods are not available. In addition, future studies may reveal if relative power will be able to show progression of the disease.

Footnotes

ACKNOWLEDGMENTS

We would like to thank the Kavli Charitable Trust in Bergen, Norway, which funded this EEG study.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Reiman

(2014) Alzheimer’s disease and other dementias: Advances in 2013. Lancet Neurol 13, 3–5.

Dubois

, Feldman

, Jacova

, Cummings

, Dekosky

, Barberger-Gateau

, Delacourte

, Frisoni

, Fox

, Galasko

, Gauthier

, Hampel

, Jicha

, Meguro

, O’Brien

, Pasquier

, Robert

, Rossor

, Salloway

, Sarazin

, de Souza

, Stern

, Visser

, Scheltens

(2010) Revising the definition of Alzheimer’s disease: A new lexicon. Lancet Neurol 9, 1118–1127.

Korf

, Wahlund

, Visser

, Scheltens

(2004) Medial temporal lobe atrophy on MRI predicts dementia in patients with mild cognitive impairment. Neurology 63, 94–100.

Visser

, Verhey

, Knol

, Scheltens

, Wahlund

, Freund-Levi

, Tsolaki

, Minthon

, Wallin

, Hampel

, Burger

, Pirttila

, Soininen

, Rikkert

, Verbeek

, Spiru

, Blennow

(2009) Prevalence and prognostic value of CSF markers of Alzheimer’s disease pathology in patients with subjective cognitive impairment or mild cognitive impairment in the DESCRIPA study: A prospective cohort study. Lancet Neurol 8, 619–627.

Mosconi

, Mistur

, Switalski

, Tsui

, Glodzik

, Li

, Pirraglia

, De Santi

, Reisberg

, Wisniewski

, de Leon

(2009) FDG-PET changes in brain glucose metabolism from normal cognition to pathologically verified Alzheimer’s disease. Eur J Nucl Med Mol Imaging 36, 811–822.

Resnick

, Sojkova

, Zhou

, An

, Ye

, Holt

, Dannals

, Mathis

, Klunk

, Ferrucci

, Kraut

, Wong

(2010) Longitudinal cognitive decline is associated with fibrillar amyloid-beta measured by [11C]PiB. Neurology 74, 807–815.

Lehmann

, Koenig

, Jelic

, Prichep

, John

, Wahlund

, Dodge

, Dierks

(2007) Application and comparison of classification algorithms for recognition of Alzheimer’s disease in electrical brain activity (EEG). J Neurosci Methods 161, 342–350.

Bennys

, Rondouin

, Vergnes

, Touchon

(2001) Diagnostic value of quantitative EEG in Alzheimer’s disease. Neurophysiol Clin 31, 153–160.

Huang

, Wahlund

, Dierks

, Julin

, Winblad

, Jelic

(2000) Discrimination of Alzheimer’s disease and mild cognitive impairment by equivalent EEG sources: A cross-sectional and longitudinal study. Clin Neurophysiol 111, 1961–1967.

10.

Claus

, Strijers

, Jonkman

, Ongerboer

de Visser BW

, Jonker

, Walstra

, Scheltens

, van

Gool WA

(1999) The diagnostic value of electroencephalography in mild senile Alzheimer’s disease. Clin Neurophysiol 110, 825–832.

11.

Ommundsen

, Engedal

, Oksengard

(2011) Validity of the quantitative EEG statistical pattern recognition method in diagnosing Alzheimer’s disease. Dement Geriatr Cogn Disord 31, 195–201.

12.

Snaedal

, Johannesson

, Gudmundsson

, Blin

, Emilsdottir

, Einarsson

, Johnsen

(2012) Diagnostic accuracy of statistical pattern recognition of electroencephalogram registration in evaluation of cognitive impairment and dementia. Dement Geriatr Cogn Disord 34, 51–60.

13.

Adler

, Brassen

, Jajcevic

(2003) EEG coherence in Alzheimer’s dementia. J Neural Transm (Vienna) 110, 1051–1058.

14.

Coben

, Danziger

, Storandt

(1985) A longitudinal EEG study of mild senile dementia of Alzheimer type: Changes at 1 year and at 2.5 years. Electroencephalogr Clin Neurophysiol 61, 101–112.

15.

Jeong

(2004) EEG dynamics in patients with Alzheimer’s disease. Clin Neurophysiol 115, 1490–1505.

16.

Roh

, Park

, Ko

, Park

, Lee

, Han

, Jo

, Yang

, Jung

(2011) Region and frequency specific changes of spectral power in Alzheimer’s disease and mild cognitive impairment. Clin Neurophysiol 122, 2169–2176.

17.

Dierks

, Ihl

, Frolich

, Maurer

(1993) Dementia of the Alzheimer type: Effects on the spontaneous EEG described by dipole sources. Psychiatry Res 50, 151–162.

18.

Duffy

, Albert

, McAnulty

(1984) Brain electrical activity in patients with presenile and senile dementia of the Alzheimer type. Ann Neurol 16, 439–448.

19.

Breslau

, Starr

, Sicotte

, Higa

, Buchsbaum

(1989) Topographic EEG changes with normal aging and SDAT. Electroencephalogr Clin Neurophysiol 72, 281–289.

20.

Ponomareva

, Korovaitseva

, Rogaev

(2008) EEG alterations in non-demented individuals related to apolipoprotein E genotype and to risk of Alzheimer disease. Neurobiol Aging 29, 819–827.

21.

Wolf

, Jelic

, Gertz

, Nordberg

, Julin

, Wahlund

(2003) A critical discussion of the role of neuroimaging in mild cognitive impairment. Acta Neurol Scand Suppl 179, 52–76.

22.

Moretti

, Babiloni

, Binetti

, Cassetta

, Dal

Forno G

, Ferreric

, Ferri

, Lanuzza

, Miniussi

, Nobili

, Rodriguez

, Salinari

, Rossini

(2004) Individual analysis of EEG frequency and band power in mild Alzheimer’s disease. Clin Neurophysiol 115, 299–308.

23.

Prichep

(2007) Quantitative EEG and electromagnetic brain imaging in aging and in the evolution of dementia. Ann N Y Acad Sci 1097, 156–167.

24.

Petersen

(2004) Mild cognitive impairment as a diagnostic entity. J Intern Med 256, 183–194.

25.

Jelic

, Johansson

, Almkvist

, Shigeta

, Julin

, Nordberg

, Winblad

, Wahlund

(2000) Quantitative electroencephalography in mild cognitive impairment: Longitudinal changes and possible prediction of Alzheimer’s disease. Neurobiol Aging 21, 533–540.

26.

Jelic

, Blomberg

, Dierks

, Basun

, Shigeta

, Julin

, Jensen

, Lannfelt

, Winblad

, Wahlund

(1998) EEG slowing and cerebrospinal fluid tau levels in patients with cognitive decline. Neuroreport 9, 157–160.

27.

Kramberger

, Kareholt

, Andersson

, Winblad

, Eriksdotter

, Jelic

(2013) Association between EEG abnormalities and CSF biomarkers in a memory clinic cohort. Dement Geriatr Cogn Disord 36, 319–328.

28.

Smailovic

, Koenig

, Kareholt

, Andersson

, Kramberger

, Winblad

, Jelic

(2017) Quantitative EEG power and synchronization correlate with Alzheimer’s disease CSF biomarkers. Neurobiol Aging 63, 88–95.

29.

Engedal

, Snaedal

, Hoegh

, Jelic

, Bo Andersen

, Naik

, Wahlund

, Oeksengaard

(2015) Quantitative EEG applying the statistical recognition pattern method: A useful tool in dementia diagnostic workup. Dement Geriatr Cogn Disord 40, 1–12.

30.

Braekhus

, Ulstein

, Wyller

, Engedal

(2011) The Memory Clinic–outpatient assessment when dementia is suspected. Tidsskr Nor Laegeforen 131, 2254–2257.

31.

Winblad

, Palmer

, Kivipelto

, Jelic

, Fratiglioni

, Wahlund

, Nordberg

, Backman

, Albert

, Almkvist

, Arai

, Basun

, Blennow

, de Leon

, DeCarli

, Erkinjuntti

, Giacobini

, Graff

, Hardy

, Jack

, Jorm

, Ritchie

, van

Duijn C

, Visser

, Petersen

(2004) Mild cognitive impairment–beyond controversies, towards a consensus: Report of the International Working Group on Mild Cognitive Impairment. J Intern Med 256, 240–246.

32.

Delorme

, Makeig

(2004) EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods 134, 9–21.

33.

Oostenveld

, Fries

, Maris

, Schoffelen

(2011) FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell Neurosci 2011, 156869.

34.

Lee

, Girolami

, Sejnowski

(1999) Independent component analysis using an extended infomax algorithm for mixed subgaussian and supergaussian sources. Neural Comput 11, 417–441.

35.

Whitham

, Lewis

, Pope

, Fitzgibbon

, Clark

, Loveless

, DeLosAngeles

, Wallace

, Broberg

, Willoughby

(2008) Thinking activates EMG in scalp electrical recordings. Clin Neurophysiol 119, 1166–1175.

36.

Whitham

, Pope

, Fitzgibbon

, Lewis

, Clark

, Loveless

, Broberg

, Wallace

, DeLosAngeles

, Lillie

, Hardy

, Fronsko

, Pulbrook

, Willoughby

(2007) Scalp electrical recording during paralysis: Quantitative evidence that EEG frequencies above 20 Hz are contaminated by EMG. Clin Neurophysiol 118, 1877–1888.

37.

Widmann

, Schroger

(2012) Filter effects and filter artifacts in the analysis of electrophysiological data. Front Psychol 3, 233.

38.

Iaccarino

, Singer

, Martorell

, Rudenko

, Gao

, Gillingham

, Mathys

, Seo

, Kritskiy

, Abdurrob

, Adaikkan

, Canter

, Rueda

, Brown

, Boyden

, Tsai

(2016) Gamma frequency entrainment attenuates amyloid load and modifies microglia. Nature 540, 230–235.

39.

Barry

, Clarke

, Johnstone

, Magee

, Rushby

(2007) EEG differences between eyes-closed and eyes-open resting conditions. Clin Neurophysiol 118, 2765–2773.

40.

Chen

, Feng

, Zhao

, Yin

, Wang

(2008) EEG default mode network in the human brain: Spectral regional field powers. Neuroimage 41, 561–574.

41.

Laufs

(2008) Endogenous brain oscillations and related networks detected by surface EEG-combined fMRI. Hum Brain Mapp 29, 762–769.

42.

Fox

, Wolfson

, Ranck

Jr (1986) Hippocampal theta rhythm and the firing of neurons in walking and urethane anesthetized rats. Exp Brain Res 62, 495–508.

43.

Mitchell

, Ranck

Jr (1980) Generation of theta rhythm in medial entorhinal cortex of freely moving rats. Brain Res 189, 49–66.

44.

Zhang

, Jacobs

(2015) Traveling theta waves in the human hippocampus. J Neurosci 35, 12477–12487.

45.

Brier

, Thomas

, Snyder

, Benzinger

, Zhang

, Raichle

, Holtzman

, Morris

, Ances

(2012) Loss of intranetwork and internetwork resting state functional connections with Alzheimer’s disease progression. J Neurosci 32, 8890–8899.

46.

Rossini

, Del Percio

, Pasqualetti

, Cassetta

, Binetti

, Dal Forno

, Ferreri

, Frisoni

, Chiovenda

, Miniussi

, Parisi

, Tombini

, Vecchio

, Babiloni

(2006) Conversion from mild cognitive impairment to Alzheimer’s disease is predicted by sources and coherence of brain electroencephalography rhythms. Neuroscience 143, 793–803.

47.

Jack

Jr , Wiste

, Lesnick

, Weigand

, Knopman

, Vemuri

, Pankratz

, Senjem

, Gunter

, Mielke

, Lowe

, Boeve

, Petersen

(2013) Brain beta-amyloid load approaches a plateau. Neurology 80, 890–896.

48.

Babiloni

, Lizio

, Del Percio

, Marzano

, Soricelli

, Salvatore

, Ferri

, Cosentino

, Tedeschi

, Montella

, Marino

, De Salvo

, Rodriguez

, Nobili

, Vernieri

, Ursini

, Mundi

, Richardson

, Frisoni

, Rossini

(2013) Cortical sources of resting state EEG rhythms are sensitive to the progression of early stage Alzheimer’s disease. J Alzheimers Dis 34, 1015–1035.

49.

Babiloni

, Triggiani

, Lizio

, Cordone

, Tattoli

, Bevilacqua

, Soricelli

, Ferri

, Nobili

, Gesualdo

, Millan-Calenti

, Bujan

, Tortelli

, Cardinali

, Barulli

, Giannini

, Spagnolo

, Armenise

, Buenza

, Scianatico

, Logroscino

, Frisoni

, Del Percio

(2016) Classification of single normal and alzheimer’s disease individuals from cortical sources of resting state EEG rhythms. Front Neurosci 10, 47.

50.

Frisoni

, Bocchetta

, Chetelat

, Rabinovici

, de Leon

, Kaye

, Reiman

, Scheltens

, Barkhof

, Black

, Brooks

, Carrillo

, Fox

, Herholz

, Nordberg

, Jack

Jr , Jagust

, Johnson

, Rowe

, Sperling

, Thies

, Wahlund

, Weiner

, Pasqualetti

, Decarli

(2013) Imaging markers for Alzheimer disease: Which vs how. Neurology 81, 487–500.