Obstructive Sleep Apnea is one of the most common types of sleep-disordered breathing (SDB) (1–3). It is characterized by recurrent upper airway collapse during sleep, often leading to repeated awakening, sleep fragmentation, and reduced slow-wave sleep (4, 5). Increasingly, OSA is regarded as a systemic disease that impacts multiple organs of the metabolic, cardio- and cerebrovascular systems, elevating all-cause mortality (6, 7).

Sleep fragmentation and frequent awakening in OSA patients lead to daytime sleepiness, reduced working and learning efficiency, cognitive dysfunction, as well as increased risks of diseases such as hypertension, stroke, diabetes, and thus reduced overall quality of life (8). At present, several major problems remain unsolved regarding the pathophysiology as well as the clinical diagnosis and treatment of OSA: (1) clinicians face significant difficulties in evaluating the disease burden of OSA due to its highly heterogeneous clinical manifestations (9), with symptoms ranging from morning fatigue to memory impairment (see Table 1 for the lack of inter-group differences in reported symptoms); (2) Polysomnography is the gold standard for OSA diagnosis (10, 11), but the current measures of disease severity (AHI, ODI, the proportions of each sleep stage) do not strongly and consistently relate to comorbidity and mortality risks (12–14); and (3) the current suite of clinical assessment questionnaires on sleep quality, daytime sleepiness, and cognitive function, viz. the Epworth Sleepiness Scale (ESS), Pittsburgh Sleep Quality Index (PSQI), Mini-Mental State Examination and Montreal Cognitive Assessment (MoCA), tends to be subjective, and lacks specificity and sensitivity in the early stages of the disease. Therefore, current clinical practice calls for better biomarkers that might objectively capture the disease burden faced by patients to better guide therapeutic decisions.

Power spectral analysis (PSA) provides a more objective alternative perspective for the analysis of EEG signals by leveraging Fast Fourier Transform (FFT) to convert time-domain EEG signals (amplitude vs. time) to frequency-domain signals (amplitude vs. frequency). This technique reveals additional information about brain activities during sleep, which might otherwise be hidden to clinicians examining the plain EEG signals on PSG tracings. PSA quantifies the amplitude of each frequency component band that constitutes the EEG waveform in the form of a power spectrum, with amplitude plotted against frequency (15). The literature has much on the physiological significance of each frequency band during different sleep stages: (1) delta activities (1–4 Hz) is generally considered to be an indicator of sleep depth and homeostasis during NREM sleep (16, 17); (2) theta activities (4–8 Hz) in the frontal region serve as a marker of emotional memory consolidation (18, 19); (3) alpha activities (8–12 Hz) are suggested to be a cortical signature of visual perception and mental imagery (20), and a reflection of neuronal activities, with low alpha levels indicating a state of excitation and vice versa (21, 22); (4) sigma activities (12–15 Hz) during N2 sleep serve as a proxy for the sleep spindle and is shown to facilitate learning and memory consolidation (23) and protect sleep from external disturbances (24, 25); and (5) high-frequency activities in the beta (15–30 Hz) and gamma (30–40 Hz) ranges are associated with arousals (24, 26, 27). As such, PSA enables a closer examination of the relative changes in prominence of each frequency band in OSA patients, to reveal the functional significance that underlies the change in EEG patterns.

In this study, we applied the PSA technique to analyze the EEG power spectral distribution variations between OSA patients and primary snoring patients during NREM and REM sleep. Moreover, clinical variables were examined for their correlation with the observed between-group EEG power differences, so as to reveal pathological manifestations of the underlying EEG changes and the clinical significance thereof.

This study aimed to characterize EEG power in different frequency bands during REM sleep and NREM sleep in patients with OSA. By computing the normalized spectral power densities, we found elevated normalized PSD in the delta band and decreased normalized PSD in the other frequency bands in the severe OSA group compared to the other two groups in NREM sleep. In REM sleep, similar changes were observed in the frontal region.

Two aspects distinguish this study from the previously published works on EEG spectral analysis in OSA patients: (1) previous studies have mostly focused their investigation on the EEG spectral distribution during wakefulness in OSA patients, with results showing increased low-frequency activity (delta and theta) in frontal and central regions compared with healthy controls (35, 36) and (2) few studies have so far analyzed the relative changes in the power of each frequency band in OSA patients during sleep. One previous study by Appleton et al. (37) demonstrated the absolute band spectral power in all frequency bands was higher for patients with higher AHI, and no changes were shown in the relative proportion or importance of each frequency band [see also a study by Kang et al. (38) showing similar increases in the beta and delta frequency bands].

This study thus set the focus on EEG recordings during sleep and adopted a relative measure of EEG band spectral power by normalizing the total EEG power for each participant for cross-participant comparisons, to allow a closer examination of the relative contributions by each frequency band to the total power, and how the severity of OSA mediates this change. Our results showed elevated delta power spectral density in the severe OSA group during the N2 stage compared to both the snoring only group and the mild-moderate OSA group. Correspondingly, reduced power spectral density was observed in other frequency bands in the severe OSA group compared to the other two groups. This demonstrates a shift in relative power contribution from the higher frequencies to the lower frequencies, i.e., the delta band, in the severe OSA patients. This is in agreement with previous EEG studies on OSA patients showing an increase in delta power during sleep (38, 39) and wakefulness (34). The methodology of computing normalized EEG power, however, allowed this paper to provide a more detailed analysis of the change in the relative importance of each power band, specifically, showing the relatively reduced spectral power in faster frequencies in NREM sleep, which previous studies failed to reveal.

There are a few potential explanations for this increase in delta power. In line with the two key pathophysiological mechanisms of OSA patients, hypoxia and sleep fragmentation, correlation analysis was conducted to examine whether variations in delta spectral power could be predicted by clinical variables associated with the two characteristics. For the former, hypoxia, our analysis showed that delta EEG spectral power was negatively correlated with MinSaO₂ and positively correlated with T90%. That is, OSA patients who suffered from more severe hypoxia during the night exhibited higher spectral power in the delta frequency band. This is supported by previous studies, which demonstrated an increase in delta spectral power (40, 41) and a decrease in beta power (38) in healthy volunteers under hypoxia conditions [see however a mouse study showing contrary results (42)]. In terms of sleep fragmentation, it was found that frontal delta spectral power in NREM sleep was positively correlated with the patients' arousal index and negatively correlated with the proportion of time in N3 sleep. That is, patients with higher EEG slowing are more prone to arousals, resulting in sleep fragmentation and a reduction of deep sleep. However, it has to be cautiously noted that the evidence provided in this study is purely correlational. Future studies are required to investigate the exact causal relationship between hypoxia, sleep fragmentation, and EEG slowing.

Clinically, changes in EEG power spectral distributions and especially the EEG slowing could entail significant prognostic consequences. Studies have reported greater EEG slowing in association with poorer performance in psychomotor vigilance tasks (41) and driving simulators (43, 44). Another observational study has reported EEG slowing in frontal lateral regions in patients with memory difficulties and mild cognitive impairment (45). It is also worth noting that this increased delta band activity was observed in frontal, central, and occipital regions in the NREM stage, as was also observed previously (44). This corroborates with existing brain imaging studies demonstrating cortical activation and compromised gray and white matter integrity across multiple brain regions (frontal and parietal cortex, temporal lobe, anterior cingulate, hippocampus, and cerebellum) (46, 47). This may partly account for the wide variability of neurobehavioral disorders in OSA patients (46).

This thus suggests that the changes in slow-wave activities may be one of the pathological mechanisms of OSA. Consistent with previous studies that demonstrated pathological EEG changes in the frontal area (46), we found a significant increase in delta spectral power and a decrease in spectral power in the faster frequency bands in severe OSA patients during both NREM and REM sleep in the frontal lobe. Given that the frontal lobe is believed to be primarily responsible for regulating higher cognitive functions such as goal-directed executive function, attention, motivation, and memory, these impairments in higher cognitive functions are frequently reported in OSA patients (48). Moreover, this is made worse by the fact that the frontal lobe is more sensitive to sleep fragmentation or hypoxic damage than other brain regions (49). In our study, no apparent differences were observed among the three groups in terms of patient complaints including memory function decline, daytime sleepiness, and insomnia; neither did the PSQI questionnaires reveal any inter-group distinctions. In contrast, significant changes in the power density spectrum were detected in multiple brain regions, sensitive enough even in our set of young and middle-aged patients (39.8 ± 10.1 years). Therefore, EEG spectral analysis could be a potentially useful technique to signal early pathological brain changes in OSA patients.

There are a few limitations in this study. Our cohort consists only of young and middle-aged male patients, making our findings potentially less generalizable to the overall OSA population. Another limitation is that we chose to only analyze stage N2 among the three NREM stages, though N2 sleep is usually the most stable and it accounts for the largest proportion (~45–55%) of the total sleep time in adults. Lastly, the control group chosen in this study is primary snoring patients instead of healthy individuals, partly due to the retrospective and hospital-based nature of the study.

In summary, in OSA patients, abnormal activations of the cerebral cortex, that is, the increase in slow-wave EEG activities and the decrease in fast-wave activities, represent an increase in disease severity, degree of hypoxia, sleep fragmentation, and repeated awakening. This study also demonstrated the utility of PSA as a tool in EEG analysis in complementing the traditional PSG parameters to assess disease severity and progression.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by the Research Ethics Committee of the Second Affiliated Hospital of Soochow University, Suzhou, China (JD-LK-2018-004-02). The patients/participants provided their written informed consent to participate in this study. Informed consent was sought from all patients involved, with the understanding that anonymized data could be used for research and publications.

SL collected and analyzed data and drafted the manuscript. JS designed the study, co-drafted the manuscript, was involved in the clinical diagnosis, and management of patients. YL analyzed data and co-drafted the manuscript. JinW conducted the medical assessment of the patients. QW analyzed the PSG data for the study. JX and JiaW provided guidance on experimental design and statistical analyses. RC designed experiments, participated in coordination, and critically revised the manuscript. All authors read and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.