Waking EEG Cortical Markers of Chronic Pain and Sleepiness

Abstract

Objective

Spectral power analyses of EEG recordings are reported to distinguish the cortical activity of individuals with chronic pain from those of controls. Further study of these spectral patterns may provide a greater understanding of the processes associated with chronic pain, in addition to providing potential biometric markers of chronic pain for use in both clinical and research settings. However, sleep deprived groups have demonstrated similar characteristics in their spectral power characteristics, particularly in alpha bandwidth power activity.

Methods

103 individuals with chronic pain provided resting awake EEG data in addition to ratings of pain and sleep quality. Two Principal Axis Factor analyses using Promax rotation produced one pain and one sleep factor from relevant questionnaire data provided by participants. These factors were then used to test hypothesized relationships with alpha and theta bandwidth power at the frontal and parietal areas of the cortex.

Results

Our findings suggest that reductions in alpha bandwidth power are independently associated with both chronic pain intensity ratings and measures of sleep deficits. Conversely, theta bandwidth power was not found to be associated with either chronic pain or sleep quality measures.

Conclusions

This study’s findings support that chronic pain intensity and sleep deficits are related to the Alpha spectral bandwidth activity in individuals with chronic pain.

Introduction

Electroencephalography (EEG) and magnetoencephalography (MEEG) bandwidth spectral power characteristics associated with chronic pain have been reported in a number of studies [1–5]. Those reports assist us in understanding how brain activity might change in the presence of pain and how altered brain activity might contribute to the pain experience. Spectral power characteristics also have the potential for use as a highly sought after biometric marker of chronic pain, for example, for use with individuals who cannot otherwise communicate their experience [6] and in research where the potential for assessment bias needs to be minimized [7]. However, the validity of using EEG spectral power as a metric for chronic pain has yet to be fully established.

One issue that has not been addressed is the potential biasing impact of sleep deficits on EEG-measured activity. Sleep deficits are commonly reported by people with chronic pain [8], and the EEG bandwidth markers associated with sleep deprivation have similarities to those associated with chronic pain [3,9–14]. These findings raise important basic questions: 1) Can we distinguish between the contributions of chronic pain and sleep deficits found within the EEG bandwidth power spectrum? 2) If so, what are the independent contributions of chronic pain and sleep deficits to the EEG markers identified? And finally, 3) can these markers be viewed as viable metrics in the study of chronic pain?

Chronic Pain and EEG

The presence of chronic pain is associated with the ongoing activation of a number of pain-related [15], although not necessarily pain-specific [16], central networks. Ongoing pain can be associated with neuroplastic changes within these networks, which can consolidate and facilitate pain processing and may further result in the perception of pain independent of peripheral neural activation [17]. Although pain-related neuroplastic changes in a number of structures and networks have been identified, those that occur in the thalamus are likely to be a key feature in a number of chronic pain conditions [18]. The thalamus acts as a gate or “relay station” for the majority of sensory input to the cortex from the periphery and also receives stimulation and information from cortical pyramidal cells [19]. Thus, the thalamus is known to play an important role in sensory processing and perception. Neuroplastic changes associated with ongoing pain may therefore play a role in the hyperpolarization of thalamic relay cells [20] and thereby contribute to some types of chronic pain disorders.

Thalamic hyperpolarization among individuals with chronic pain is posited to explain the detection of low-frequency resonances in the theta range (4–8 Hz) [1–5] localized in the thalamus [1] and high theta (6–9 Hz) activity [2–5,21] localized in the posterior insular, inferior posterior parietal cortex [2], somatosensory cortex [5,21], and frontal areas [4]. Further, low theta (4–6 Hz) activity has been reported to be localized in the parieto-temporal cortex [2]. Alpha band activity is thought to represent inhibitory drive from activation of GABAergic synapses from the thalamic projections to the cortex [22]. Alpha (8–12 Hz) activity is also reported to be lower in individuals with chronic pain [9], localized in the insula [2] and frontal electrodes [9], with reduced high alpha (13–15 Hz) activity reported in the frontal electrodes [3]. This pattern of findings indicating abnormal low frequency oscillations between the thalamus and the cortex has been labeled “thalamo-cortical dysrhythmia” [1,23] and may underlie, or at least put individuals at risk for, the experience of pain.

In summary, increased theta activity is commonly observed in individuals with chronic pain relative to those without chronic pain and is generally localized in cortical sites that are known to be associated with chronic pain, such as the insula, posterior parietal cortex, and the somatosensory cortex [24]. In addition, reduced alpha bandwidth activity is also reportedly lower in individuals with chronic pain, localized in the insula and particularly in the frontal lobes, when compared with healthy controls [2,9].

Sleep and EEG

Sleep complaints are present in up to 88% of chronic pain disorders [8], with more than 50% of insomnia patients also presenting with chronic pain [25]. Notably, sleep deficits are also associated with secondary outcomes, such as enhanced pain intensity [26–29] and reduced pain thresholds [30]. While the impact of pain on sleep can be understood in terms of arousal-provoking cognitions that restrict sleep onset [31] and perceptual stimuli that promote awakenings from sleep [32], the cortical mechanisms associated with sleep deficits and their ability to modulate pain intensity is unclear.

One mechanism reported to mediate the relationship between poor sleep and subsequent pain is negative mood [33]. Cortical structures involved in the processing of emotionally salient information usually include the amygdala, which in turn is influenced by a variety of connected systems, particularly the medial-prefrontal cortex [34]. Interestingly, amygdala activation in response to negative emotional stimuli is greater in sleep-deprived conditions than it is in individuals who are not sleep deprived [34,35]. Negative mood is associated with both pain [36,37] and sleep deficits [38], so it is not surprising that there are comparable cortical components involved. Cortical structures that are involved in pain and its affective processing include the periaqueductal grey, amygdala, anterior cingulate cortex, and anterior insula [36].

The interpretation of EEG data related to sleep onset is largely based on cortical activity within the alpha band frequency (8–12 Hz) [39]. In summary, studies that report the resting wake EEG characteristics of patients with sleep deficits have shown that in an eyes-closed condition, alpha (8–12 Hz) activity decreased as individuals got sleepier [10–14], especially in frontal locations [13]. In addition, more theta activity was shown to be associated with sleep deprivation [12,13], also maximal in frontal areas [13, 14]. Correlations between the amount of alpha and theta in the eyes-closed condition have also been found to be low, suggesting that independent processes underlie these two oscillations [14].

To date, only one study has measured waking EEG power spectral characteristics of patients with chronic pain and measures of sleep quality. Jensen etal. (2013) examined 13 individuals with spinal cord injury and chronic pain and reported, after a series of three different neurofeedback protocols (12 sessions total), that participants had modest decreases in worst pain and pain unpleasantness scores. Pre- to post-treatment changes in EEG bandwidth activity were also reported with a decrease in theta power, an increase in alpha power, but no change in beta power. Sleep quality was assessed using the six-item medical outcomes study (MOSS) short-form scale [40]. While no significant treatment-related changes were found in measures assessing sleep quality or fatigue, there was an unexplained nonsignificant trend for an increase in fatigue from pre- to post-treatment that continued through the three months of follow-up [41].

Pain and Sleep Summary

The impacts of both chronic pain and sleepiness on EEG-measured cortical activity have notable similarities and differences. The similarities include the prominence of theta and alpha activity; in particular, both conditions are associated with an increase in theta activity and reduced alpha activity. However, increased theta activity in the cortex is localized in the parietal lobe/somatosensory cortex in individuals with chronic pain, whereas increased theta activity is localized in the frontal areas in individuals with sleepiness. Individuals with either chronic pain or sleepiness are found to have reduced alpha activity (relative to those without pain or who are less sleepy) maximally localized in the frontal areas of the cortex. Finally, studies on chronic pain indicate a relationship between theta and alpha activity, whereas in sleepiness alpha activity seems to be unrelated to theta activity.

Purpose of the Current Study

Given this background, the aims of the current study are to: 1) evaluate whether we can distinguish between the contributions of chronic pain and sleep deficits found within the EEG bandwidth power spectrum; 2) measure the independent contributions of the chronic pain and sleep deficits to the EEG markers identified; and 3) determine the extent to which these markers may be viewed as useful measures in the study of chronic pain. Based on the available findings, as well as current thinking regarding the effects of sleepiness on the waking EEG power spectrum, we hypothesized that higher levels of reported sleepiness would be associated with lower levels of alpha activity. We also hypothesized that higher levels of sleepiness would be associated with higher levels of theta activity. Given that some of the findings regarding the maximal localization of both decreased alpha activity and increased theta activity are preliminary (found in only one or two studies so far), we also wanted to test the following hypotheses, which we viewed as more exploratory: 1) lower alpha activity related to increased sleepiness would be maximally found located in in activity measures by electrodes in frontal areas; and 2) higher levels of theta activity associated with increased sleepiness would be found in activity measured by electrodes placed in frontal areas (see Figure 1).

Figure 1.

Hypothesized cortical markers associated with pain intensity and sleep quality in individuals with chronic pain.

With regard to the relationship between chronic pain intensity and the waking EEG power spectrum, we hypothesized that more chronic pain intensity would be associated with higher levels of theta activity. To the extent that previous findings regarding the maximal locations of increased theta activity and chronic pain intensity and that the associations between reduced alpha activity and chronic pain intensity are limited to only a few studies, we further wished to test the following hypotheses, which we also viewed as more exploratory: 1) more pain intensity associated with increased theta activity would be maximally located in brain activity measures from electrodes over the parietal areas; 2) more pain intensity would be related to less alpha activity; and 3) the lower levels of alpha activity associated with chronic pain intensity would be found maximally using activity measured from electrodes placed over frontal areas (see Figure 1). Finally, given that no one to our knowledge has yet examined the independent contributions of both sleepiness and chronic pain intensity to the waking EEG power spectrum, we sought to explore the associations between these conditions and their relative (and independent) contributions to alpha and theta activity.

Methods

Participants

One hundred three individuals (56 spinal cord injury [SCI], 16 with multiple sclerosis [MS], 15 muscular dystrophy [MD], one acquired amputation [AMP], and 15 low back pain [LBP]) age 24 to 81 years (mean = 54.1 years, SD = 13.6 years; male/female = 43/60) with chronic pain took part in this study. The participants were recruited primarily through a registry of individuals with physical disabilities who have expressed an interest in participating in research studies. The registry is managed by the Department of Rehabilitation Medicine at the University of Washington in Seattle, Washington, USA. This registry currently contains 348 individuals with MS and 516 individuals with SCI. One hundred fifteen of those with MS and 264 of those with SCI met the criteria for participating in the current study. A second source for the participants was an ongoing survey funded by the National Institute on Disability and Rehabilitation Research study that recruits individuals with MS or SCI. The third source for participants in this study was the University of Washington Medical Centre Clinics where posters and brochures describing the study were posted. Ethical review board approval from the University of Washington’s School of Medicine was obtained for this study.

Inclusion and Exclusion Criteria

Inclusion criteria included: 18 years of age or older, a rating of 4 or higher on a 0–10 numerical rating scale of average pain intensity in the last week, experienced chronic pain for six months or more, and able to read, speak, and understand English. Exclusion criteria included: severe cognitive impairments, defined as one or more errors on the six-item screener [42], a psychiatric condition or psychiatric symptoms that would interfere with participation (specifically active suicidal ideation with intent to harm oneself or active delusional or psychotic thinking), a history of a medical condition that could produce an abnormal EEG and interfere with the tests of the effects of treatment on EEG (e.g., epilepsy, history of traumatic brain injury that involved a loss of consciousness), a chronic pain problem that predated the SCI, MS, MD, AMP, or LBP diagnosis (e.g., painful diabetic neuropathy, predisability chronic headache), any medical condition that required immediate medical treatment as determined during the medical screening evaluation (e.g., evidence for a possible previously undiagnosed cancer by the study physicians during the initial medical evaluation).

EEG Assessment Procedures

Participant Preparation

After briefings about methods and providing consent, participants were fitted with a 128-channel hydrocell net connected to a GES 300 high-density EEG acquisition system (EGI, Eugene, OR, USA). Scalp impedances were measured and maintained below 50 kohm. Subjects were seated comfortably and coached to release tension from scalp, face, neck, and shoulders and to sit still and quiet with eyes closed during periods of EEG recording.

Recording Sequence

Participants were given a short introductory pain questionnaire followed by a six-minute EEG recording. Then, during a two-minute break, a second pain questionnaire was administered. This was followed by a second six-minute EEG recording and a final pain questionnaire.

Sleep Onset Monitored During Recording

During EEG data collection, the technician queried the participant if they suspected that the participant was dosing off. The EEG data were also examined for evidence indicating sleep onset during the EEG assessment—<50% Alpha with vertex waves signifying N1 sleep, specifically sleep spindles, or K-complexes with a plan to eliminate from analyses any records suggesting that the participants were sleeping during the EEG. However, no evidence for sleep onset was found within the EEG of any of the 103 participants, indicating that the study procedures were effective in collecting only waking EEG activity.

Bad Channel Replacement

Raw recordings were clipped down to two five-minute intervals following the pain questionnaire that occurred in the middle of the recording session. Channels located below the ears and on the face were excluded from subsequent analysis. Time sequences from included channels were detrended by subtracting the best-fitting cubic spline and then filtered to include frequency components between 0.1 and 55 Hz. Each channel's time sequence was compared with the time sequence obtained from the mean of the channel's nearest neighbors. If the correlation coefficient, R, between a channel and the mean of its neighbors fell below 0.5, the channel was marked as bad. The R values for the remaining channels were then recalculated, excluding the bad channel from all the other channels' collections of neighbors, and the process was repeated until there were no R values below 0.5. Data from each bad channel was then replaced by the mean of its nearest (“non-bad”) neighbors. Channels were then re-referenced to average.

Regions of Interest

Regions of interest (ROIs) were centered on sensors located at F3, F4, CP3, and CP4. Each ROI included the central sensor, the nearest neighbors of those central sensors, and the nearest neighbors of those nearest neighbors. All subsequent artefact screening and spectral analysis was performed on the average of the sensors included in each ROI.

Artefact Exclusion

Motion and EMG artefacts were detected using a robust outlier criterion based on nonparametric z-scores (NPZ) [43]. A sample was deemed to contain artefact if its NPZ score (with respect to the entire sequence) exceeded a criterion appropriate for the specific type of artefact. To identify motion artefacts resulting from mechanical disruption of sensor contact, an ideal filter was applied with pass band spanning 0.1 to 55 Hz. The section of the sequence collected during the two-minute questionnaire period midway through the recording session was discarded. A time point was deemed to contain motion artefact if its NPZ score with respect to its entire filtered sequence was greater than 4.75 (∼99th percentile as estimated from a Monte-Carlo simulation). Time spans not containing these large deflections were subdivided into two-second-long intervals, which were thereafter designated “stationary intervals.” EMG values for each stationary interval were estimated by summing the Fourier amplitudes for frequency components between 65 and 90 Hz. An interval was deemed to contain EMG artefact if the NPZ score with respect to all the stationary interval EMG values was greater than 3 (∼95th percentile). Spectra from intervals not containing EMG artefact were used to compute an average spectrum for the recording session.

Spectral Analysis

The sum of power (microvolts squared = μV²) was computed for theta (4 < 8 Hz) and alpha (8 < 12 Hz) bands. These values were also converted to relative power, that is, the proportion of summed power in the band divided by the total power of the entire spectrum (from 0.5 to 125 Hz, excluding 60 and 120 Hz). Mean power density estimates were log10-transformed to normalize the distribution [44].

Pain Measures

Pain intensity ratings were obtained using 0–10 (0 = “no pain,” 10 = “pain as bad as you can imagine”) numerical rating scales three times. Current pain was assessed midway through the 10-minute EEG assessment and again at the end of the assessment. Least, worst, and average pain “in the past five minutes” ratings were obtained at the midway and end of EEG session assessment points.

Sleep and Sleep Quality Measures

Sleep quality was assessed using the following questions asked prior to the EEG assessment. Questions 1, 2, and 3 were modified from questions from the Pittsburgh Sleep Quality Index (PSQI) [45], whereas question 4 was taken from the Brief Pain Inventory (BPI) [46]. Both the PSQI and the BPI are well established as valid and reliable measures.

1. How would you rate the quality of the sleep you got last night overall? The participant was given the response options of: 1 = very good, 2 = fairly good, 3 = fairly bad, 4 = very bad.

2. How many hours did you sleep last night? The participant was to provide a number of hours slept.

3. When did you wake up today? Time: ___:___ (military)

4. In the past week, how much has pain interfered with sleep? The participants were then to respond using a 0–10 scale, where a 0 means that “pain does not interfere with that activity” and a 10 means that “pain completely interferes.”

Data Analysis

All data analyses were performed using IBM SPSS Statistics for Windows (version 21) [47]. We began by producing a correlation matrix of pain measures and also of sleep measures to assess them for factor analysis. Factor analysis was selected as a preliminary procedure as it is a statistical method used to describe variability among observed, correlated variables to potentially reduce data sets to a lower number of unobserved latent variables [48]. Among the various types of factor analyses available, principal axis factor was the preferred option as it is an extraction method used for understanding the shared variance of the variables. The goal in using principal axis factor analysis is factor structure interpretation and data reduction, as opposed to the goal for principle component analysis, which is usually only data reduction [48].

Version 21 of IBM SPSS offers five rotation methods: Varimax, Direct Oblimin, Quartimax, Equamax, and Promax. Three of those are orthogonal (Varimax, Quartimax, and Equimax), and two are oblique (Direct Oblimin and Promax). Tabachnick and Fidell (2007, p. 646) argue,

The best way to decide between orthogonal and oblique rotation is to request oblique rotation (e.g., Direct Oblimin or Promax from SPSS) with the desired number of factors and look at the correlations among factors. If factor correlations are not driven by the data, the solution remains nearly orthogonal. If correlations exceed 0.32, then there is 10% (or more) overlap in variance among factors, and enough variance to warrant oblique rotation. [48]

Following the factor analyses, we examined the association between demographic and descriptive variables (age, sex, type of pain [SCI, MS, MD, AMP, or LBP]) and the pain and sleep factors to determine if they needed to be controlled for in subsequent analyses.

In order to test our hypotheses regarding sleep and pain factors and their relationship to alpha and theta bandwidth spectral power activity and their maximal source location, initial analyses were undertaken using Pearson’s r correlations. The correlation results were then used to determine whether a mediation analysis was to be undertaken. As sleep characteristics had been reported to affect pain characteristics [26–29] and pain to affect sleep quality [31–33], it was determined that if the sleep factor, pain factor, and target bandwidth power were all found to be correlated to each other, with at least at a weak relationship of 0.29 [49], then a mediation analysis could be performed.

On the basis that both pain and sleep variables are hypothesised to be independently associated with reduced alpha activity localized in the frontal lobes, we performed a moderation analysis to examine if there was an interaction effect from pain and sleep factors on alpha bandwidth power. Mediation and moderation analyses were undertaken using an add-on for SPSS called “PROCESS” [50]. PROCESS uses an ordinary least squares or logistic regression–based path analytic framework for estimating direct and indirect effects in simple and multiple mediator and moderation models. Bootstrap and Monte Carlo confidence intervals are implemented for inference about indirect effects, including various measures of effect size. (All analysis protocols were submitted and registered on the Body in Mind group webpage prior to undertaking analyses [51].)

Results

Absolute and relative alpha and theta bandwidth measures for electrode sites F3, F4, CP3, and CP4 are presented in Table 1 as a contribution for future meta-analyses.

Table 1.

Summary table of the alpha and theta spectral bandwidths in absolute and relative formats

N = 103	F3 Mean (SD)	F4 Mean (SD)	CP3 Mean (SD)	CP4 Mean (SD)
Absolute bandwidth
Alpha standard (8 < 12 Hz)	9.98 (11.74)	10.23 (12.06)	6.38 (9.41)	6.41 (8.96)
Theta standard (4 < 8 Hz)	6.13 (6.90)	6.80 (7.40)	3.75 (4.70)	4.03 (6.31)
Relative bandwidth log10 (μV2/Hz)
Alpha standard (8 < 12 Hz)	0.20 (0.11)	0.20 (0.10)	0.21 (0.10)	0.21 (0.10)
Theta standard (4 < 8 Hz)	0.13 (0.06)	0.14 (0.06)	0.14 (0.06)	0.14 (0.06)

Pain Factor

The pain questions fulfilled the criteria of sample size for factor analysis [52]. We further used Tabachnick and Fidell’s recommendation of inspecting the correlation matrix (often termed “factorability of R”) for correlation coefficients over 0.32 [48] and Hair etal.’s (1995) rule of thumb measure of ±0.30 = minimal, ±0.40 = important, and ±0.50 = practically significant [53]. The data also fulfilled the Kaiser-Meyer-Olkin measure of sampling adequacy (KMO = 0.921) and the Bartlett’s Test of Sphericity chi-square was 1,756.35 (df = 28, sig. < 0.001). A factor analysis was undertaken using principal axis with a goal of reducing the number of pain variables; further, we utilized a Promax rotation because correlations between variables exceeded 0.32 [48], as per the rationale outlined in the Data Analysis section.

The first three Eigen values were 7.41, 0.23, and 0.16, and as only one factor fulfilled the criteria of an Eigen value greater than 1.0, we employed only that factor, which we labeled “pain intensity” for subsequent analyses. The factor score coefficients were current pain midway through the EEG assessment (r = 0.44), current pain at the end of the assessment (r = −0.12), average pain midway through the assessment (r = 0.55), average pain at the end of the assessment (r = −0.05), least pain midway through the assessment (r = 0.00), least pain at the end of the assessment (r = −0.20), worst pain midway through the experiment (r = 0.27), and worst pain at the end of the assessment (r = 0.09). The total percentage of the variance accounted for by pain intensity was very large at 93%.

Sleep Factor

Three out of the four sleep quality questions fulfilled the criteria of sample size for factor analysis [52], Tabachnick and Fidell’s recommendation for correlation coefficients over 0.30 [48] and Hair etal.’s (1995) rule of thumb measure [53]. The responses to the question “What time did you wake up today?” did not fulfil these criteria and were excluded from further analyses. The remaining three sleep variables also fulfilled the Kaiser-Meyer-Olkin measure of sampling adequacy (KMO = 0.610) and the Bartlett’s Test of Sphericity chi-square was 68.15 (df = 3, sig. < 0.001). We undertook a factor analysis of these variables for the extraction of principal axis factor with a Promax rotation as the sleep data fulfilled the criteria outlined by Tabachnick and Fidell (2006) that was also utilized for our pain factor [48].

The first three Eigen values were 1.90, 0.74, and 0.36, and as only one factor fulfilled the criteria of an Eigen value greater than 1.0, we employed only that factor for subsequent analyses. We labeled the first factor “poor sleep quality” because the factor score coefficients indicated that it was strongly associated with hours slept the previous night (r = −0.49) and quality of sleep (r = 0.44), and very weakly associated with the extent to which pain interfered with sleep in the last week (r = 0.10). The total percentage of the variance accounted for by poor sleep quality was 63%. Preliminary analyses found no significant association between age, sex, or type of injury (SCI, MS, MD, AMP, or LBP) and the factors of poor sleep quality, pain intensity, alpha bandwidth power, or theta bandwidth power. Therefore, no demographic variables were controlled for in the subsequent analyses.

Associations Between Alpha and Theta Activity and Pain and Sleep Factors

The correlations between alpha bandwidth power, theta bandwidth power, and the pain and sleep factors are presented in Table 2. As can be seen, pain intensity was found to have negative, moderate associations with alpha bandwidth power from electrodes at sites F3, F4, CP3, and CP4. Pain Intensity was also found to have no apparent association with theta bandwidth power at F3, F4, CP3, and CP4. Poor sleep quality was found to have significant, negative, moderate associations with alpha bandwidth power from electrodes at sites F3, F4, and CP4. Poor sleep quality was also found to have no apparent association with theta bandwidth power at F3, F4, CP3, and CP4 (see Figure 2).

Table 2.

Correlation matrix of pain intensity and poor sleep quality on alpha and theta spectral power bandwidths recorded at F3, F4, CP3, and CP4 electrode sites

N = 103	Pain intensity				Poor sleep quality
Relative bandwidth Log10 (μV2/Hz)	F3	F4	CP3	CP4	F3	F4	CP3	CP4
Alpha standard (8 < 12 Hz)	−0.25*	−0.27**	−0.22*	−0.30*	−0.24*	−0.26**	−0.19	−0.23*
Theta standard (4 < 8 Hz)	−0.04	0.01	0.02	0.01	0.04	0.00	0.08	0.04

^*P < 0.05, **P < 0.01.

Figure 2.

Observed cortical markers associated with pain intensity and sleep quality in individuals with chronic pain.

As pain can affect sleep quality, and conversely sleep had been reported to influence pain ratings, we had considered the possibility of a mediation effect between these variables and our hypothesized cortical relationships. This possibility was enhanced with our hypothesized findings that both factors of pain intensity and poor sleep quality had negative, moderate associations with alpha bandwidth power in the frontal areas. However, pain intensity was not found to be significantly associated with poor sleep quality (r = 0.11, P > 0.05); hence no mediation effect between poor sleep quality, pain intensity, and alpha bandwidth power in the current sample is possible.

We proceeded with moderation analyses to explore whether there was an interaction between pain intensity and poor sleep quality on their relationship with reduced alpha bandwidth power measured by F3 and F4 electrodes in the frontal areas of the cortex (see Table 3). As can be seen, no significant interaction effect was evident. Moreover, the interaction effect size of our combined factors was inconsequential at less than 1%. This indicates that while the relationship of pain intensity on reduced alpha bandwidth power was found to be of weak to moderate strength, its impact on reduced alpha bandwidth in frontal areas was predominately independent and its contribution did not increase or reduce the influence of poor sleep quality. This finding also indicates that while the relationship of poor sleep quality on reduced alpha bandwidth power was found to be of moderate strength, its impact on reduced alpha bandwidth in the frontal areas was also predominately independent and its contributions did not increase or reduce the influence of pain intensity.

Table 3.

Moderation analyses of the associations between pain intensity, poor sleep quality, and alpha bandwidth power measured by the frontal and parietal electrodes

N = 103	F3	F4	CP3	CP4
Pain intensity	ΔR2 = 0.0056,	ΔR2 = 0.0040,	ΔR2 = 0.0005	ΔR2 = 0.0025
X	F(1, 99) = 0.6307	F(1, 99) = 0.4516	F(1, 99) = 0.0519	F(1, 99) = 0.2811
Poor sleep quality interaction effect, P	0.43	0.50	0.82	0.60

ΔR² = R-square increase due to interaction.

Post Hoc Analyses

Based on our finding that both pain intensity and poor sleep quality had a weak to moderate association with alpha bandwidth activity recorded at the CP3 and CP4 electrodes in the parietal areas, we proceeded with a moderation analysis to explore whether there was an interaction effect between pain intensity and poor sleep quality and reduced alpha bandwidth power at the CP3 and CP4 electrode sites (see Table 3). As can be seen, no significant interaction effect was evident. Moreover, the interaction effect size of our combined factors was negligible at less than 0.3%. This indicates that while the relationship of pain intensity on reduced alpha bandwidth power was found to be of weak to moderate strength, its impact on reduced alpha bandwidth in the parietal areas was largely independent and its contribution did not increase or reduce the influence of poor sleep quality. This finding also indicates that while the relationship of poor sleep quality on reduced alpha bandwidth power was found to be of moderate strength, its impact on reduced alpha bandwidth in parietal areas was also essentially independent and its contributions did not increase or reduce the influence of pain intensity.

Based on our findings that both pain intensity and poor sleep quality had no apparent association with theta bandwidth power, we considered that one factor’s effect could be moderating the other. This further prompted us to proceed with additional moderation analyses to explore whether there was an interaction effect between pain intensity and poor sleep quality on theta bandwidth activity (see Table 4). As can be seen, no significant interaction effect between pain intensity and poor sleep quality was evident at theta bandwidth activity recorded at both frontal and parietal sites. In addition, the interaction effect size of our combined factors was negligible at less than 1.5%. This indicates that while no relationship of pain intensity on theta bandwidth power was found, it also did not increase or reduce the influence of poor sleep quality’s impact on theta bandwidth in the frontal and parietal areas. These findings also indicate that while no relationship of poor sleep quality on theta bandwidth power was found, its impact on theta bandwidth in frontal and parietal areas was also independent and did not increase or reduce the influence of pain intensity.

Table 4.

Moderation analyses of the associations between pain intensity, poor sleep quality, and theta bandwidth power measured by the frontal and parietal electrodes

N = 103	F3	F4	CP3	CP4
Pain intensity	ΔR2 = 0.0009	ΔR2 = 0.0047	ΔR2 = 0.0099	ΔR2 = 0.0011
X	F(1, 99) = 0.0919	F(1, 99) = 0.4630	F(1, 99) = 0.3293	F(1, 99) = 0.1106
Poor sleep quality interaction effect, P	0.76	0.50	0.80	0.74

ΔR² = R-square increase due to interaction.

Discussion

This study found that both pain intensity and sleep quality were related to reductions in alpha bandwidth power in the frontal areas of the cortex, hence supporting previously noted studies [2,9–14]. However, pain intensity and sleep quality were also unexpectedly found to be related to reductions in alpha bandwidth power in the parietal areas. Furthermore, our findings concerning theta bandwidth power did not support our hypotheses. Specifically, we had hypothesized that pain intensity would have a positive relationship with theta bandwidth power maximal in the parietal lobes; however, we found no discernible relationship between these variables at the electrode sites measured [12,13]. The unexpected lack of a relationship between pain intensity and theta power activity in this study, therefore, provided no support for the role of theta activity in the model for thalamo-cortical dysrhythmia in chronic pain patients [1–5]. Moreover, our hypothesis that poor sleep quality would have a positive relationship with theta bandwidth power that was maximal in the frontal areas was also not supported [12–14].

Despite our mixed findings, this study was still able to address the primary questions posited. The first question that this study endeavored to answer was “Can we distinguish between the contributions of chronic pain and sleep deficits found within the EEG bandwidth power spectrum?” Our findings indicate that the impact of pain and sleep deficits on EEG outcomes were distinguishable from each other in that both conditions were found to be associated with discrete EEG characteristics using relatively simple and noninvasive measures. Moreover, our findings suggest that there was no cumulative or inhibitory effect of pain intensity on sleep quality, or vice versa, with regard to EEG outcomes. This lack of an interaction effect suggests that despite pain and sleep deficits having some similarities with regard to reduced alpha bandwidth power, their individual contributions were likely to be unique [54].

The second question that we sought to address was “What are the independent contributions of the chronic pain and sleep deficits to the EEG markers identified?” In terms of effect size, the independent contributions of pain intensity on alpha bandwidth power were reasonably similar to those of sleep deficits. Pain ratings and sleep deficits were both found to have measurable impact on EEG spectral characteristics, with each factor accountable for approximately 10% of the reduction in alpha power in the frontal and parietal areas. However, the contributions of pain and sleep deficits to theta bandwidth power were indiscernible using our methodological approach and could not be evaluated further.

The third question we sought to address was “Can these markers be viewed as viable metrics in the study of chronic pain?” Our tentative response to this query is that our findings demonstrate that EEG bandwidth power may be a useful indicator of some chronic pain or sleep deficit conditions. Further study is clearly needed, however, to determine if the patterns we observed can be replicated in future research, not least because of the discrepancies between our findings and the very few studies previously undertaken in this area. However, if our aim is also to identify brain activity patterns that can be targeted with neuromodulatory treatments such as neurofeedback, hypnosis, and meditation (treatments known to result in changes in brain activity) [55–57] or specific patterns of brain activity found to be associated with, and that possibly underlie, the pain experience, the alpha power outcomes referred to in this study could be targeted for change or as a way to potentially influence or give patients more control over their pain [55–57].

There are a number of limitations of the study that should be considered when interpreting these results. First, we did not include any questions to assess current (daytime) sleepiness, nor did we include any objective measures of sleep behavior such as actigraphy or polysomnography. It could also have been advantageous to include more long-term measures of chronic pain and sleep deficit characteristics as there may be differences between the short-term and long-term relationships of these conditions on EEG outcomes. In addition, even though our sample size was comparable with previous studies of this nature, future research would benefit fom a larger sample of participants with chronic pain or from including participants with higher average pain intensities. These design improvements would increase the power (ability to detect true effects) and the reliability of potential outcomes.

Our findings raise important questions for future research. First, it would be useful to understand why some aspects of our results differed from previous studies. It is possible, for example, that differences emerged because different chronic pain conditions may result in different EEG profiles. In addition, our findings point to the need for research to help understand why there is a reduction in alpha bandwidth power in the frontal lobes of the cortex in both chronic pain and sleep deficit conditions. While there is some research associating reduced frontal alpha power with increased attention [39], our finding does not fully enhance our understanding of the impact of chronic pain and poor sleep on cognitive functioning.

In conclusion, this study provides support that both pain intensity and sleep quality are associated with alpha band power in individuals with chronic pain. Considering the impact of sleep deficits on enhancing pain perception, our results further indicate that EEG markers may be a useful metric with patients where both chronic pain and sleep deficits are likely to be present.