Running Title: EEG Biomarkers of Neurobehavioural Functioning in OSA

Title: Quantitative Electroencephalogram Measures in Adult Obstructive Sleep Apnea - Potential Biomarkers of Neurobehavioural Functioning

Running Title: EEG Biomarkers of Neurobehavioural Functioning in OSA

Authors: Angela L. D’Rozario1,2,3, Nathan E. Cross*1, Andrew Vakulin1,4 Delwyn J. Bartlett1,5, Keith K.H Wong1,3,5, David Wang1,3,5, Ronald R. Grunstein1,3,5

*co-first author

Affiliations: 1. CIRUS Centre for Sleep and Chronobiology, Woolcock Institute of Medical Research, The University of Sydney, Sydney, NSW, Australia. 2. School of Psychology, Faculty of Science, Brain and Mind Centre and Charles Perkins Centre, The University of Sydney. 3. Department of Respiratory and Sleep Medicine, Royal Prince Alfred Hospital & Sydney Local Health District, Sydney, NSW, Australia. 4. Adelaide Institute for Sleep Health: A Flinders Centre of Research Excellence, School of Medicine, Faculty of Medicine, Nursing and Health Sciences, Flinders University, Bedford Park, South Australia, Australia. 5. Sydney Medical School, The University of Sydney, NSW, Australia.

Corresponding author: Angela D’Rozario, Woolcock Institute of Medical Research, PO Box M77, Missenden Road, Camperdown, 2050, NSW, Australia. Ph: +612 9114 0435 Fax: +612 9114 0010Email:

Conflict of interest: The authors do not have any conflicts of interest to disclose.

Acknowledgments: This work was supported by NeuroSleep, Centre of Research Excellence in Interdisciplinary Sleep Health funded by the Australian National Health & Medical Research Council (NHRMC, APP1060992), and CIRUS (APP571421), Centre for Sleep and Chronobiology at the Woolcock Institute of Medial Research. A Dora Lush Priority NHMRC Scholarship (ALD), Australian Postgraduate Award (NEC) and Australian NHMRC Research Fellowships (RRG, AV, DW) funded researchers who undertook this review.

Summary

Obstructive sleep apnea(OSA) results in significantly impaired cognitive functioning and increased daytime sleepiness in some patients leading to increased risk of motor vehicle and workplace accidents and reduced productivity. Clinicians often face difficulty in identifying which patients are at risk of neurobehavioural dysfunction due to wide inter-individual variability, and disparity between symptoms and conventional metrics of disease severity such as the apnea hypopnea index. Quantitative electroencephalogram (EEG) measures aredeterminants of awake neurobehavioural function in healthy subjects. However, the potential value of quantitative EEG (qEEG) measurements as biomarkers of neurobehavioural function in patients with OSA has not been examined. This review summarises the existing literature examining qEEG in OSA patients including changes in brain activity during wake and sleep states, in relation to daytime sleepiness, cognitive impairment and OSA treatment. It will speculate on the mechanisms which may underlie changes in EEG activity and discuss the potential utility of qEEG as a clinically useful predictor of neurobehavioural function in OSA.

Key Words: Quantitative EEG analysis, power spectral analysis, signal processing, sleep disordered breathing, cognition, performance

Abbreviations

ABGArterial blood gas

AHIApnea hypopnea index

COPDChronic obstructive pulmonary disease

CPAPContinuous positive airway pressure

DFADetrended fluctuation analysis

DS%Deep sleep percentage

EEGElectroencephalogram/electroencephalography

ESSEpworth sleepiness scale

FFTfast Fourier transform

KSSKarolinska sleepiness scale

MSLTMultiple sleep latency test

MWTMaintenance of wakefulness test

NREMNon-rapid eye movement

ODIOxygen desaturation index

OSAObstructive sleep apnea

PFCPrefrontal cortex

PSAPower spectral analysis

PVTPsychomotor vigilance task

qEEGQuantitative electroencephalography

REMRapid eye movement

RCTRandomized control trial

SaO2Arterial oxyhaemoglobin saturation

SWASlow wave activity

SWSSlow wave sleep

Introduction

Obstructive sleep apnea (OSA) is a frequently under-recognised sleep breathing disorder [1], and is highly prevalent affecting almost a quarter of middle-aged women and half of middle-aged men in the general community [2], with increasing prevalence tied to the rise in obesity [3]. Untreated OSA is associated with increased daytime sleepiness, problems with attention and cognitive functioning, and increased risk of motor vehicle and workplace accidents[4]. Thisis a significant problem in high-risk populations such as in transport workers where vigilance failure has serious safety consequences[5, 6]. Importantly, there is wide inter-individual variability in the risk of vigilance failure as not all OSA patients are sleepy or show functional impairments [7]. Moreover, sleepiness in transport workers may be underestimated [8]. Current clinical metrics of disease severity (apnea hypopnea index, AHI and oxygen saturation indices) do not strongly and consistently relate to functional impairment [9-11], leaving physicians in a quandary when assessing patients regarding driving risk and need for treatment.

Electroencephalogram (EEG) recordings can be quantitatively analysed revealing variations in brain activity which are modulated by homeostatic and circadian processes, these two distinct biological mechanisms and their interactions are modelled by the two-process model of sleep-wake regulation[12][13]. Circadian regulation(process C) of the sleep-wake cycle is influenced by the intrinsic biological clock located in the suprachiasmatic nucleus of the hypothalamus,while the homeostatic process S represents sleep propensity or ‘sleep drive’ that intensifies with increasing time awake and dissipates with time spent asleep. Previous work in healthy subjects has shown relationships between quantitative EEG characteristics of NREM sleep and daytime performance and sleep propensity highlighting the potential value of EEG measures as biomarkers of sleepiness and cognitivefunctionin OSA[14]. However therelationships between EEG activity and daytime sleepiness and cognitive impairment in OSA have not been well elucidated, a gap addressed with this review.

EEG characteristics of wake and sleep

Thalamocortical networks are largely responsible for the distinct patterns of neuronal firing on the EEG which differentiates wake from distinct stages of sleep[15]. Wakefulness and REM sleep exhibit similar EEG features, characterised by low-voltage fast-frequency ‘desynchronised’activity.The EEG characteristics of NREM sleep varies significantly with sleep stage. N2 sleep is characterised by K-complexes (slow oscillations <1 cycle per second, Hz) and sleep spindles (waxing and waning oscillations of 0.5-3 seconds, usually within the frequency range of11–16 Hz) and N3 or SWS is dominated by high-voltage, slow wave activity (SWA) in the delta (typically 0.5–4 .5Hz) frequency range.

EEG power spectral analysis

The sinusoidal waveforms of the EEG are often defined by their frequency i.e. waveforms in the theta band have frequencies of 4.5 - 8 Hz; alpha waves have a frequency of 8 – 12 Hz, etc. In reality, the EEG recorded during wake or sleep is often a complex mixture of different waveforms superimposed on one another. The complex patterns of the EEG can be created by adding together simpler sine waves of different frequencies, amplitudes and phase relationships. Conversely, the EEG can be deconstructed into waveforms with different frequency components. Power spectral analysis (PSA) based on fast Fourier transform (FFT) is the most common method for processing the EEG signal in the current literature. The EEG is transformed from the time domain (amplitude versus time) to the frequency domain (amplitude versus frequency)[16]. This process allows for greater insight into brain activity during sleep which would otherwise be masked when using conventional sleep staging criteria.PSA quantifies the amplitude and prevalence of the different frequency sine waves, which compose the EEG waveform[17]. The amplitude or magnitude of the components for a specified frequency is then plotted as a power spectrum with frequency on the x-axis and amplitude on the y-axis and is expressed in mean square values (absolute power)representing the contribution of that particular frequency to the composite EEG waveform[17]. Relative EEG power (often called power density) is defined as the absolute power of a given frequency band divided by the total absolute power across all frequencies. Relative power provides a calculation of the contribution of EEG activity within a given frequency range based on an individual’s distinct EEG composition.

Quantitative EEG measures as determinants of neurobehavioural function in healthy adults

In healthy people, resting wake EEG power varies with increasing time awake[18-20]; shows circadian variation [21, 22]; and relates to measures of performance and sleepiness [19, 23, 24].Slowing of the EEG is reflected in increasedactivity in the delta, theta and alpha frequency rangesduring prolonged periods of wakefulness. This EEG slowing has beenlinked to greater subjectively-rated levels of sleepiness and fatigue[18, 24]and objectively-measured lapses in attention and vigilance failure[25, 26].EEG frequencies are altered with circadian timing, and these correspond with changes in cognitive processing [22, 27, 28].

Sleep EEG activity also relates to sleepiness and cognitive performance. In young healthy subjects, reduced theta activity in non-rapid eye movement (NREM) sleep predictsincreaseddaytime sleepiness measured by the multiple sleep latency test (MSLT)[29], while greater SWA during slow wave sleep (SWS)relates to improved declarative memory, procedural learning andfaster reaction times[30]. A similar relationship was found between pre-frontal cortex (PFC) generated SWA during sleep and awake neurobehavioural function in healthy older adults [31]. Research also supports the functional role of sleep spindles in sleep-dependent memory processes with increased sleep spindle density and sigma (12-16 Hz) EEG power related to improved learning and memory consolidation[32]. Reductions in these sleep EEG features in older adults have been implicated in the weakening of sleep-dependent learning and memory processes observed in ageing[33].

The relationship between EEG and neurobehavioural function described in healthy adultsunderscores the potential clinical value of EEG-based candidate biomarkers in OSA. EEG biomarkers may accurately and reliably phenotype patient groups cross-sectionally to indicate disease/non-disease states or predict treatment or other future outcomes. While a growing number of studies have examined qEEG activity in OSA, the reported findings have not been well summarised. The review describes differences in brain activity during sleep and awake states in OSA patients and controls. It further examines how abnormal EEG profiles in OSA manifest, and the utility of qEEG measures as candidate biomarkers of neurobehavioral dysfunction.

Method

An electronic search using PubMed and Scopus databases was conducted. We included original research papers published up to December 2015. The keywords utilised in the systematic search were ‘EEG spectral power’, ‘EEG spectral analysis’, ‘EEG power’, ‘quantitative EEG’, ‘EEG topography’, ‘EEG spectra’, and these were combined with either ‘obstructive sleep apnea’, ‘sleep disordered breathing’, or ‘continuous positive airway pressure’.

The search was limited to include full-text and English language publications. References of the selected studies were reviewed to identify additional relevant studies that were not found in the initial search. We included all observational studies (n=2), case-controlled studies (n=13) and continuous positive airway pressure (CPAP) studies (n=9) which quantitatively analysed EEG data sampled during wake (eyes open and eyes closed states) and sleep (NREM and REM sleep from polysomnography (PSG) recordings) using power spectral analysis in adults with OSA to evaluate qEEG activity as markers of daytime sleepiness and cognitive performance, and/or differences in OSA and controls.We excluded any publications including participants under 18 years of age or with significant co-morbidities (e.g. stroke, epilepsy).

Quantitative EEG measures during wakefulness in OSA

This section examines the evidence for EEG-based measures during wakefulness to discriminate OSA patients from healthy controls, and how waking EEG activity relates to neurobehavioural functioning focusing on the common symptoms of daytime sleepiness and cognitive performance. Furthermore, we examine the effects of CPAP treatment on waking EEG profiles to critically evaluate the potential of qEEG as biomarkers of neurobehavioural function (daytime sleepiness and cognitive performance).

Differentiating OSA patients from controls using wake qEEG measures

Seven studies[34-40] have directly compared differences in awake EEG spectral power between OSA patients and healthy subjects, see Table 1. In OSA patients, increasedEEG activity in slower frequencies during wakefulness was found in 6 of the 7 studies, predominantly in the frontal and central regions [34, 36-40], while one study showed no between group differences in EEG spectral power [35]. EEG slowing during wake in OSA was reflected by higher absolute and relative EEG powerwithin slower frequency bands(delta and theta) in five studies [34, 37-40],while four studies reported a higher ratio of slow to fast frequency EEG activity (EEG slowing ratio = (delta+theta) /(alpha+beta)absolute power), see Table 1. These findings were robust even when controlling for age [36]. One study also reported an increasein faster frequency (absolute beta)EEG power during sustained wakefulness, which possibly reflects a greater cortical resource requirement to maintain wakefulness as a compensatory mechanism [37].

Changes in waking EEG power are inconsistently related to PSG metrics of OSA severity, see Table 1. In severe OSA patients AHI, ODI and minimum SaO2 were not related to delta power but were positively correlated to alpha power whilst theta and beta power correlated with worse hypoxaemia[41]. In another study of severe OSA patients, a greater EEG slowing ratio was associated with higher arousal index (frontal, central, temporal, parietal and occipital regions: all r between 0.41 and 0.48, all p < 0.05) and higher AHI (parietal region: r=0.44, p=0.028, trends for all other cortical regions) [36]. In a more diverse group of mild to severe OSA patients, increased relative delta and theta power were related to worse hypoxemia [38]. A difference in absolute waking EEG power was only observed between severe OSA patients and controls in this study, suggesting altered waking EEG activity is dependent on OSA severity [38]. Other studies have failed to find any association between PSG-derived OSA severity measures and waking EEG power[34, 40, 42, 43].

Overall, these observations strongly suggest there is a general slowing of the EEG, observed as greater power in the slower frequency bands, during wakefulness in OSA patients compared with controls. It is less clear how these EEG changes relate to PSG-derived metrics of sleep-disordered breathing.

Relationship between wake qEEG and daytime sleepiness in OSA

Seven studies examined the relationship between wake qEEG and daytime sleepiness in OSA patients [36-40, 42, 44], with inconsistent results. Three of these studies [38, 39, 44] assessed self-rated sleepiness and found a significant relationship with PSA-derived qEEG measures during wakefulness, see Table 1. Greater sleepiness (higher Epworth sleepiness scale (ESS) score) was correlated with lower relative alpha power and higher relative delta and theta power across all brain regions [38]. Another study identified a relationship between greater sleepiness (higher Karolinska Sleepiness Scale (KSS) score) and higher relative delta, theta and alpha power during eyes open and higher relative delta and alpha power during eyes closed at the central derivation [39]. Similarly in a third study, a randomized control trial (RCT) design investigating the effects of modafinil in CPAP withdrawal, found a higher KSS score was significantly correlated with higher awake alpha/delta ratio, fast ratio (alpha+beta)/(delta+theta) in OSA patients during CPAP withdrawal [44]. In contrast, measures of EEG slowing during wake failed to differentiate sleepy from non-sleepy patients using the ESS at a cut-off score of 10 [42]. Using the MSLT or maintenance of wakefulness test (MWT) to objectively measure daytime sleepiness in OSA patients, no significant relationships were found with waking EEG measures and sleepiness [34, 40, 42].

These inconsistent findings between sleepiness measures and waking EEG may be explained by the small studies, differences in subjective and objective tests, as well as the timing of the administration of the test (i.e. the influence of circadian or homeostatic factors). The condition in which the EEG was recorded e.g. eyes open, eyes closed or averaged across both conditions, and whether the spectral power were presented as absolute or relative values or as a ratio of slow to fast frequency activity may also explain discrepant findings.

Overall, increased slow frequency EEG activity (delta, theta ranges) during wakefulness appears to correlate with greater subjective sleepiness, yet there are no established relationships between qEEG and objectively-measured sleepiness to date. This may be due to limitations in how these tests measure and define sleepiness; and the lack of a gold-standard test for measuring sleepiness in OSA. Though the MWT is promoted as an objective test of daytime sleepiness and treatment efficacy, its utility to accurately identify OSA patients at risk of neurobehavioural dysfunction is limited [45]. More studies with larger numbers and consistent methodologies are required to investigate the complex relationships of the waking EEG and daytime sleepiness in OSA. This in turn will determine whether wake qEEG slowing is a candidate biomarker for sleepiness in OSA and can differentiate a subset of patients with a high risk- ‘alertness failure’ phenotype.

Relationship between wake qEEG and cognitive performance in OSA

Increased theta and delta EEG activity, and decreased beta activity during wake are associated with attentional performance lapses and vigilance failure in healthy subjects [23, 25, 46, 47]. To date, there is limited direct evidence examining links between the waking brain activity and cognitive performance in OSA [36, 39, 44]. Waking qEEG (eyes closed) was not associated with attention using the four choice reaction time test in OSA patients while associations between wake EEG and performance were not assessed in the control group [36]. In contrast, a relationship was found between brain activity and performance on a psychomotor vigilance task (PVT) and a driving simulator task in both OSA and healthy volunteers across 40 hours of extended wakefulness [39]. Poorer driving performance and vigilance was correlated with higher delta, theta and alpha EEG power (eyes closed) and increased theta power (eyes open) [39], see Table 1. In addition to PSA, this study used a technique for analysing non-stationary physiological data called detrended fluctuation analysis (DFA)[48, 49]. The DFA method quantifies fluctuations in the EEG and provides a single measure called the scaling exponent. The DFA scaling exponent of the resting wake EEG measured at baseline predicted impaired driving after 24-hrawake in OSA whilst conventional PSA measures did not. Both standard and novel signal processing techniques are needed to confirm the utility of qEEG biomarkers. In a recent study employing similar techniques, significant associations were reported between increased awake EEG alpha/delta ratio and poorer driving simulator and PVT performance in 23 OSA patients during a CPAP withdrawal RCT cross-over study [44]. Furthermore, a higher DFA scaling exponent positively correlated with PVT lapses, and negatively correlated with PVT reciprocal reaction time [44].