Quantitative electroencephalography in cerebral amyloid angiopathy

.


Introduction
Cerebral amyloid angiopathy (CAA) is a common small vessel disease characterized by the deposition of amyloid beta within the walls of small to medium-sized cerebral blood vessels (Wermer and Greenberg, 2018).Sporadic CAA (sCAA) is estimated to account for up to 20% of spontaneous intracerebral hemorrhages (ICH) in the elderly (Biffi and Greenberg, 2011).CAA is also a major contributor to cognitive decline and a cause of transient focal neurologic episodes (TFNE) (Wermer and Greenberg, 2018).Dutchtype hereditary CAA (D-CAA) is a monogenetic variant of CAA caused by a mutation in the APP gene that has an earlier onset and faster progression than sCAA, but is otherwise pathologically and clinically similar (Bornebroek et al., 1996;Maat-Schieman et al., 1996).The clinical course for patients with sCAA and D-CAA remains unpredictable and reliable biomarkers for disease progression and particularly for cognitive decline are lacking.
In the past decades, quantitative electroencephalography (qEEG) has been increasingly used to investigate a vast array of pathologies (Fingelkurts and Fingelkurts, 2022).EEG in general is a non-invasive, well-tolerated and inexpensive method to measure brain function.It relies less on the motivation of the participant than neuropsychological testing and can be performed even in the setting of severe cognitive dysfunction or focal neurological deficits that interfere with speech or motor skills.Qualitative (visual) assessment of EEGs, however, relies heavily on the experience of the investigator and can be both time-consuming and subject to significant interrater variability (Thatcher, 2010).qEEG eliminates these factors by mathematically processing and analyzing the digitally recorded EEG, achieving a higher reliability and clinical sensitivity than visual examination for various psychiatric and neurological disorders (Thatcher, 2010).qEEG has been shown to aid in differentiating between neurodegenerative conditions such as Alzheimer's disease and dementia with Lewy Bodies (Livinț Popa et al., 2021;Chatzikonstantinou et al., 2021), to predict the conversion from mild cognitive impairment to dementia (Engedal et al., 2020), to correlate with cognitive impairment and nonmotor disease severity in Parkinson's disease (Geraedts et al., 2018a, Geraedts et al., 2018b, Zimmermann et al., 2015), and to predict outcome and early cognitive decline in stroke patients (Brito et al., 2021;Doerrfuss et al., 2020;Bentes et al., 2018).Until now, qEEG has not been used to assess patients with CAA.A correlation between qEEG measures and cognition could show the value of qEEG as a potential biomarker for cognitive dysfunction in CAA.
Thus far, only qualitative EEG results of patients with CAA have been published in case reports, a case series and one retrospective study.These EEGs were usually acquired in a clinical setting due to apparent TFNEs.Most EEGs were normal, although occasionally intermittent focal slowing was found associated with the presence of cortical superficial siderosis (cSS; Li et al., 2019;Ni et al., 2015, Viguier et al., 2018).cSS is caused by deposits of hemosiderin in the subpial layers of the brain and is associated with TFNEs, possibly through induction of cortical spreading depolarizations (Smith et al., 2021).cSS is also an important marker for future ICH (Charidimou et al., 2013).In addition to focal slowing, focal epileptic activity has been described in some cases (Mattavelli et al., 2021).A recent retrospective study showed slow and epileptic discharge activity on qualitative EEG assessment in the majority of patients with CAA, both with and without clinical epilepsy (Tabaee Damavandi et al., 2023).Linking cSS to focal slowing on qEEG would confirm the findings of the small sample of qualitative EEGs that have reported this phenomenon, as well as provide insight in how CAA in general and cSS in particular lead to cortical dysfunction.
In this paper, we investigated several qEEG measures in a cohort of patients with sCAA and D-CAA.First, we aimed to determine whether qEEG measures correlate with cognitive performance in CAA.Second, we explored whether the presence of cortical superficial siderosis is associated with (focal) slowing on qEEG.

Data availability Statement
Further information about the dataset is available from the corresponding author upon reasonable request.

Participants
For this cross-sectional study, participants were included who were taking part in two ongoing, natural history studies of the Leiden University Medical Center (LUMC) on CAA: the AURORA (nAt-URal histORy Angiopathy) study for D-CAA and the FOCAS study for sCAA.Both studies share the exact same study protocol and recruitment started in 2018; data for this study was collected in the second year of follow-up, between 2019 and 2023.Patients were recruited via the (outpatient) clinic of the LUMC.Inclusion criteria for D-CAA participants were: age !18 years and presence of the causal APP mutation or a history of symptomatic ICH (sICH) on CT/MRI suspect for CAA and !1 first-degree relative with D-CAA.Both pre-symptomatic mutation-carriers and symptomatic (defined as a history of symptomatic ICH) patients with D-CAA were included.Patients with sCAA were included if they fulfilled the criteria for probable CAA according to the Boston criteria 2.0 and had no family history of D-CAA (Charidimou et al., 2022).
For all participants, data on demographics, medical history, and clinical symptoms including history of symptomatic ICH were prospectively obtained via questionnaires on the same day as a neuropsychological assessment, EEG recording and an MRI of the brain.Level of education was subdivided into 'low', 'middle' and 'high', based on the seven categories of the Dutch Verhage score (Rijnen et al., 2020).A score of 1-4 (any education below finishing average-level secondary education) was considered low, a score of 5 (finished average-level secondary education) was categorized as middle and a score of 6-7 was considered high (finished high level secondary education or obtained a university degree).Education level was missing in 5 participants; this data was imputed as 'low', as this was the most prevalent category.

Neuropsychological assessment
The neuropsychological assessment included multiple cognitive tests, from which we selected the Montreal Cognitive Assessment (MoCA) as an overall measure of cognitive function in our study (Nasreddine et al., 2005).CAA is associated with dysfunction in multiple cognitive domains, with a similar profile to vascular dementia (Case et al., 2016).We chose to use the MoCA as it utilizes multiple cognitive domains (memory, executive functioning, attention, language, visuospatial and orientation) in its composite score and is regularly administered in clinical practice.
The studies were approved by the local ethics review boards of the medical center, and written informed consent was obtained from all participants.

EEG recording, pre-processing and analysis
EEGs were recorded under standardized conditions, with patients awake in resting condition in a supine position with the eyes closed, according to a previously described protocol (Geraedts et al., 2018b).An EEG technician monitored signal quality and potential artefacts throughout the registration and alerted patients with acoustic stimuli if drowsiness was detected on the EEG.Each EEG registration lasted a minimum of twenty minutes.If the EEG technician felt that more time was needed to obtain enough artefact free (i.e.without movement artefacts or sleep) signal for a particular participant, the registration could be lengthened by ten minutes to a total of thirty minutes.Twenty-one Ag/ AgCl EEG scalp electrodes were used in a standard 10-20 EEG electrode system.ECG and horizontal eye movement leads were added to aid the identification of artifacts.Data were acquired online using a Nihon Kohden EEG-1200 system, with a 500 Hz sampling rate, a 16-bit analog-to-digital converter, and band-filtered between 0.16 and 70 Hz.EEG data were re-referenced towards a source derivation.No further preprocessing was performed.Each EEG was visually screened by two investigators for artifacts and five unique, non-overlapping 4096-point epochs lasting 10 s with the best technical quality (i.e. with the patient awake and with the eyes closed, as well as with no or almost no movement arte-facts or artefacts of any other kind) were identified.These epochs were converted to American Standard Code for Information Interchange (ASCII) format and analyzed using BrainWave software (BrainWave version 0.9.152.4.1, C.J. Stam; available at https:// home.kpn.nl/stam7883/brainwave.html).
As this is the first study to use qEEG in the assessment of patients with CAA, we have chosen to focus mainly on spectral analysis, because these indices are most often reported in other studies and easily reproducible.Furthermore, as connectivity parameters become increasingly important in qEEG research, we have also included the phase lag index (PLI) as a reflection of functional connectivity.The PLI is thought to show the degree of synchronization between couples of signals and therefore the functional connectivity between regions of the brain; a higher PLI denotes more synchronization (Chaturvedi et al., 2019).Spectral analysis was performed by processing each epoch with a Fast Fourier Transform (FFT) in BrainWave and averaged to produce a power spectrum for each individual electrode.The frequency bands were defined as delta (0.5 to 4.0 Hz), theta (>4.0 to 8.0 Hz), alpha1 (>8.0 to 10.0 Hz), alpha2 (>10.0 to 13.0 Hz) and beta (>13.0 to 30.0 Hz).We chose to exclude gamma band power, as it is frequently strongly influenced by muscle artifacts (van Diessen et al., 2015).Relative bandpower was calculated by dividing the absolute bandpower of each frequency band by the total absolute bandpower from the FFT average per channel.We computed the relative band power for each frequency band.The average peak frequency was derived from the dominant peak (i.e. the highest absolute power) within the power spectrum of these relative band powers.Furthermore, we calculated the spectral ratio between slow (relative delta and theta) and fast (relative alpha and beta) frequencies for the whole EEG as well.As part of a post-hoc analysis, we also calculated the theta to alpha ratio (TAR).Functional connectivity was assessed by calculating the average PLI for the whole EEG as a global synchronization measure, yielding measures between 0 (indicating no coupling or coupling with a phase difference centered around 0) or 1 (indicating perfect phaselocking; Geraedts et al., 2018b;Stam et al., 2007).As a secondary analysis, we also analyzed the average PLI in the theta, alpha1 and alpha2 band; these specific bands were chosen because previous literature has most often found changes within these bands in neurodegenerative conditions (Geraedts et al., 2018b;Utianski et al., 2016;Engels et al., 2015;Chaturvedi et al., 2019;Zawis ´lak-Fornagiel et al., 2023;Yan et al., 2023).

Magnetic resonance imaging acquisition and assessment of cortical superficial siderosis
All participants underwent a 3-Tesla MRI scan of the brain (Philips Healthcare, Best, the Netherlands), using a standard 32channel head coil.For the current study, only data from the susceptibility-weighted images (SWI) was analyzed (see Supplemental Material for details on the MRI protocol).
Cortical superficial siderosis was qualitatively scored by two independent investigators.No preprocessing was performed.Presence, location, extensiveness (focal when restricted to 3 sulci, or disseminated when affecting !4 sulci), ICH relation (presence of an ICH 3 sulci) and cSS hemisphere score ((Charidimou et al., 2017b) were recorded for each participant, using to the Standards for Reporting Vascular Changes on Neuroimaging (STRIVE) criteria (Wardlaw et al., 2013).In the event of disagreement between the investigators on the score, an experienced neuroradiologist was consulted to reach consensus.

Statistics
We performed descriptive statistics to describe the baseline characteristics.The association between the MoCA, the average peak frequency, the spectral ratio and the PLI were investigated by first using univariate linear regression for each EEG measure, followed by a generalized linear model in which we included the MoCA as the dependent variable, the EEG measure and age as covariates, as well as type of condition (sCAA or D-CAA), sex and level of education as factors.As a secondary analysis, for the average peak frequency and spectral ratio, we looked at the relative bandpower of each frequency band in relation to the MoCA, while for the PLI, we looked at the theta, alpha1 and alpha2 bands.We also analyzed the correlation between the MoCA and the TAR instead of the spectral ratio within the generalized linear model.Additionally, we performed separate analyses in which we added the interaction between symptomatic hemorrhages and the average peak frequency or spectral ratio to the model, to investigate whether the correlation between EEG slowing and cognition is different between the patients who experienced a symptomatic hemorrhage and those who did not.
The average peak frequency and the spectral ratio between patients with and without siderosis, those with focal and disseminated cSS, and those with mild (cSS hemisphere score 1 or 2) and severe cSS (cSS hemisphere score 3 or 4) were compared using an independent samples T-test.In all tests, the p-value was considered significant if less than 0.05.

Patient characteristics
We included 92 patients (44 with D-CAA with a mean age of 55 years and 48 with sCAA with a mean age of 72 years, Table 1).Of these 92 participants with EEG data available, we had cognitive data for 89 participants and imaging data for 85 participants; these are the numbers included in each type (cognitive or imaging) analysis.The flowchart of inclusions is shown in Fig. 1.

qEEG measures and cognition
In the univariate analysis, a lower average peak frequency on the EEG was associated with a lower score on the MoCA (b [95% CI] = 0.986 [0.252-1.721];P = 0.009; R = 0,275; Fig. 2A): a decrease of 1 Hz in peak frequency correlated with a decrease of approximately 1 point on the MoCA.After adjustment for confounders, the correlation between average peak frequency and cognitive performance remained present (b [95%CI] = 0.935 [0.277-1.593];P = 0.005).The other covariates and factors in the model were not correlated with the MoCA (see Supplemental Material for the full models of the primary outcome).A higher spectral ratio (indicating the presence of relatively more slow activity as a proportion of the amount of fast activity within the EEG) was correlated with a lower MoCA score in the univariate analysis (b [95%CI] = À0.918[À1.761-À0.075];P = 0.033; R = 0.226; Fig. 2B), but was not statistically significant in the multivariate analysis (b [95%CI] = À0762 [À1,557-0,032]; P = 0.060).When considering the frequency bands separately in the generalized linear model in a secondary analysis, we noted that the presence of more activity in the alpha1 and alpha2 frequency was positively correlated with cognitive performance (P = 0.007 and P = 0.037), while an increased presence of theta activity was negatively correlated (P = 0.047).There was no correlation between relative beta and delta power and the MoCA.Based on these results, we conducted a post-hoc analysis in which we explored the correlation between the MoCA and the theta to alpha ratio (TAR) rather than the entire spectral ratio.We found a correlation between a lower MoCA score and a higher TAR in both the univariate (b [95%CI] = 1.213 [0.528-1.898];P = 0.001) and multivariate b ([95%CI] = 1.025 [0.390-1.660];P = 0.002) analysis.
There was no interaction between the presence of a history of sICH and the average peak frequency or the spectral ratio (average peak frequency: b À0471; P = 0.504; spectral ratio: b 0.671; P = 0.455; Fig. 3).The correlation between the qEEG and the MoCA was therefore not different between those who experienced a symptomatic intracerebral hemorrhage and those who did not.

qEEG spectral measures and cortical superficial siderosis on MRI
Fifty-one participants had some form of cSS on their MRI scan, while 34 out of a total of 85 participants did not.There was no difference between the average peak frequency (7.92 Hz vs. 8.07 Hz; P = 0.534) and spectral ratio (1.62 vs. 1.42;P = 0.307) in those with cSS compared to those without cSS (Table 2).Additionally, the average peak frequency (8.08 Hz vs. 7.89 Hz; P = 0.554) and spectral ratio (1.70 vs 1.22; P = 0.187) in those with focal and disseminated cSS was similar.The same was true for the average peak frequency (8.04 Hz vs 7.86 Hz; P = 0.549) and spectral ratio (1.33 vs 1.72; P = 0.238) in those with mild cSS compared to severe cSS.

Discussion
We found that a lower average peak frequency and a higher spectral ratio on the EEG correlated with a lower MoCA score in patients with CAA.A decrease of one Hz in the peak frequency of the EEG was correlated with a decrease of approximately one point on the MoCA.The correlation between lower average peak frequency and MoCA remained intact after correcting for type of condition, sex, age and level of education, while the spectral ratio did not.The differences in peak frequency and spectral ratio appear to be driven by alterations in alpha activity and theta activity, where the presence of more (fast) alpha activity is associated with a higher MoCA and the presence of more (slow) theta activity is associated with a lower MoCA.Based on this finding, we conducted a post-hoc analysis that showed that the TAR might show a stronger correlation to cognition in CAA than the entire spectral ratio, as it specifically utilizes the frequency bands that appear most involved.In our connectivity analysis, we found no correlation between PLI and MoCA.We also explored whether the EEG changes were related to the presence of superficial siderosis, but could not find a relationship between this measure and the qEEG.
Our findings show that there is a correlation between qEEG slowing and cognitive decline in CAA, suggesting that qEEG can be used as a biomarker for cognition in CAA.The presence of slow activity on qualitative EEG in many patients with CAA has been previously shown, but not yet been linked to cognition (Tabaee Damavandi et al., 2023).Generalized EEG slowing on qEEG and its connection to cognition has been well established in numerous other neurodegenerative diseases (Livinț Popa et al., 2021), but has not previously been investigated in CAA.Similarly to Alzheimer's disease and vascular dementia, patients with CAA with worse cognitive performance showed a lower alpha power, although it is an increase in theta power rather than delta power that additionally drives the effect in our CAA population (Livinț Popa et al., 2021).The similar findings to Alzheimer's disease are unsurprising, as Alzheimer's disease and CAA both share the accumulation of amyloid-b in their pathogenesis, although the location of this accumulation differs (Smith, 2018).There is also a significant overlap between the Alzheimer's disease and sCAA: pathology studies have demonstrated the presence of CAA in the brains of approximately 80% of patients with Alzheimer disease (Biffi andGreenberg, 2011, Kövari et al., 2013).The correlation between EEG slowing and cognitive decline appears to be mostly driven by the sCAA participants in our study, possibly reflecting the influence of concomitant Alzheimer's disease.The differences in alpha activity in patients with CAA could also be linked to the observation that CAA shows a posterior-to-anterior progression and is often most severe in the posterior brain regions, which are also the source of the alpha rhythm (Kövari et al., 2013).
Notable is that there was no substantial difference in the correlation between the average peak frequency and the spectral ratio of the EEG and the performance on the MoCA in patients with CAA with or without a symptomatic ICH.We included this analysis to explore whether the correlation between the EEG and the MoCA was driven by those participants who had experienced an sICH with subsequent parenchymal loss, but found that the correlation was not different between the participants with and without sICH.This suggests that the slowing of the EEG and the cognitive decline in our group of patients with CAA is not solely driven by incidental large hemorrhages, but possibly also by a more global process of neurodegeneration, including the slow accumulation of other vascular lesions.Other studies have similarly shown that cognitive decline in CAA occurs even without new ICHs and that patients with CAA exhibit brain atrophy in regions not directly affected by ICHs (Fotiadis et al., 2016, Smith, 2018;Charidimou et al., 2017a).Cognitive decline in CAA is now thought to be most likely caused by both specific CAA pathology comprising hemorrhages, but also ischemia (leading to microinfarction and ischemic demyelination), and concomitant AD pathology (Smith, 2018).
The benefit of using qEEG as a biomarker for cognition in CAA in conjunction with or in lieu of neuropsychological testing lies in the fact that it provides objective data that is not influenced by motivation of the participant.In the setting of more advanced cognitive decline, patients can become unwilling to participate in cognitive testing, as this can be stressful or frustrating (Kiselica et al., 2021).Similarly, neuropsychological testing can become unreliable and even futile in patients with severe dementia (Kiselica et al., 2021).In patients with CAA with focal neurological deficits due to sICH, these tests can also be influenced by impaired motor skills, neglect or aphasia.qEEG could provide additional information on cognition in these patients.Furthermore, contrary to qEEG, tests like the MoCA can be influenced by practice effects, making them less suitable for longitudinal measures (Cooley et al., 2015).These applications predominantly benefit research, however, as continued testing in the setting of severe cognitive decline often serves no clinical purpose.
We found no correlation between the PLI and the MoCA.There are no previous studies investigating PLI in patients with CAA.However, in Alzheimer's disease, the PLI had been shown to correlate to disease severity (Engels et al., 2015); it is possible that the cognitive decline in our participants was not severe enough to show significant changes in functional connectivity or that functional connectivity declines relatively late in the CAA disease process.
Qualitative EEGs performed in the clinical setting have occasionally linked focal slow activity to the presence of cSS (Li et al., 2019;Viguier et al., 2018;Ni et al., 2015).We could not confirm this finding with quantitative EEG.When comparing patients with cSS to those without cSS, we found no significant difference in peak frequency or spectral ratio.Ideally, we would have compared the EEG frequencies in lobes with cSS to lobes without cSS within the same patient, but the patients in our study had such extensive and symmetric cSS that this comparison was not possible.It could also be the case that the focal slow activity found in clinical EEGs is present only intermittently (analogous to the TFNEs related to cSS).Our study was not designed to investigate such transient phenomena.
Our study has several strengths.It is the first study that investigated qEEG in CAA and it was performed in a relatively large group of patients.Other strengths are the standardized approach to the collection of the data and the recording of the EEG, as well as the ability to look at both cognitive and radiological markers.We chose to analyze global spectral measures and average PLI; the advantage of this is the reproducibility of these measures as well as the robustness of the outcome, as this approach utilizes all electrodes and is less vulnerable to artefacts influencing single electrodes (Geraedts et al., 2018a).Furthermore, cognition is a product of the entire brain, which makes the use of overall qEEG measures appropriate.The disadvantage of using solely global measures is the inability to analyze regional differences.
Our study has other limitations.First, the MoCA is a very general measure of cognitive performance that has been shown to have a considerable ceiling effect, and cannot replace a full neuropsychological examination.Considering more cognitive tests would allow us to investigate whether EEG alterations are linked to changes in specific cognitive domains.Second, we did not include a control group, which prevents us from directly comparing patients with CAA to healthy controls or to patients with other neurodegenerative diseases, such as Alzheimer's disease.However, as our aim was to study the correlation between cognition and EEG in patients with CAA, the lack of control group does not influence our primary objective.Furthermore, differentiating patients with CAA from healthy controls is less clinical relevant for CAA, as the disease is readily diagnosed on MRI-scans or, in the case of Dutch-type hereditary CAA, via genetic testing.Third, we did not correct for multiple comparisons.In our opinion this was not necessary because we had only three primary outcome measures which were connected with each other and the other outcomes were either secondary or exploratory.However, this decision can be debated.Fourth, our cohort includes few participants with sev- ere cognitive dysfunction, as these patients are often unable to attend research days.This may have biased our findings.Fifth, although our use of five EEG epochs of ten seconds per participant allowed us to select the portion of the EEG of the highest quality, this meant that we possibly lost information in the data we did not use.Sixth, our radiological analysis only focused on exploring the effect of cSS, while CAA also has other prominent radiological markers, such as cerebral microbleeds, enlarged perivascular spaces, and specific white matter changes.We made this choice based on the clinical data linking cSS and TFNE's to focal slow activity on EEG.Seventh, we were also unable to explore regional differences between areas with and without cSS, due to the extensiveness of the cSS in our participants.Eighth, we have not yet been able to investigate qEEG as a predictive tool for cognitive worsening.Cognitive decline in CAA is often gradual and the participants in our natural history study have not yet completed enough followup visits to draw any conclusions on the utility of qEEG in predicting cognitive trajectories over time.
In conclusion, our study represents a first foray into qEEG research in CAA.Thus far, CAA has been a disease that has mainly been diagnosed and monitored by structural abnormalities on imaging.EEG allows us to look at function rather than structure and our investigation shows that qEEG can be used as a biomarker for cognition in patients with CAA.This could be beneficial for studying patients with advanced cognitive decline or those with focal deficits that influence standard neuropsychological testing.Future research could explore the correlation between qEEG and more extensive neuropsychological testing, as well as qEEG and cognition changes in longitudinal CAA cohorts in order to discover whether qEEG can be used to predict cognitive decline.More connectivity parameters as well as the gamma band could be studied in CAA.Further research into EEG findings during TFNEs or using quantified cSS measurements of certain brain regions could also shed light on the complicated relationship between transient symptoms, cSS and EEG changes.

Fig. 2 .
Fig. 2. Association of average peak frequency (A) and spectral ratio (B) with the MoCA.There is a significant correlation between a lower MoCA score and a lower average peak frequency, as well as between a lower MoCA score and higher spectral ratio.MoCA indicates Montreal Cognitive Assessment.D-CAA indicates Dutch-type cerebral amyloid angiopathy.sCAA indicates sporadic cerebral amyloid angiopathy.

Fig. 3 .
Fig. 3. Association of average peak frequency (A) and spectral ratio (B) with the MoCA for patients with and without sICH.The correlation between MoCA score and the EEG measures is not different for participants with a symptomatic intracerebral hemorrhage compared to those without a symptomatic intracerebral hemorrhage.MoCA indicates Montreal Cognitive Assessment.sICH indicates symptomatic intracerebral hemorrhage.

Table 2
Average peak frequency and spectral ratio in patients with and without cSS.