Study design.

The most frequently used type of design was cross sectional (n = 12), followed by clinical trials (n = 2) and longitudinal observational study (n = 1).

Patients’ profile.

In total, 283 individuals with chronic pain and 287 controls were evaluated. The sample size of the patient groups ranged from eight to 37 individuals. Women were more frequently assessed than men (75.3% of all studies), and five studies included only females. The age of participants, when identifiable, ranged from 19 to 63 years.

The Visual Analogue Scale (VAS) was the most frequent instrument to measure the intensity of pain. The average pain intensity was above four (range from 0 to 10) in most studies (n = 10), and it was not reported in two studies. Eleven studies collected data about pain duration, which ranged between 5.2 and 19.3 years. Only four studies used instruments to characterize the pain. These results are presented at Table 2.

Diagnosis.

The studies included participants with migraine (with and without aura), fibromyalgia, rheumatoid arthritis, osteoarthritis, low back pain, degenerative or inflammatory disc disease, neuropathic pain, radiculopathy, polyradiculopathy, polyneuritis, neuralgia, carpal tunnel syndrome, abdominal pain secondary to chronic pancreatitis, breast cancer and chronic pain of any origin. Most studies used standardized, validated and specific diagnostic criteria for chronic pain (n = 12) (Table 2).

EEG recording protocol.

Technical protocols for EEG recording varied widely among studies. According to the authors, electrode placement followed the 10–20 International System in all studies, however the number of electrodes for EEG recordings varied from four to 64, with 10 studies (66.7%) reporting more than 25 electrodes. Sampling rates ranged from 200 to 2,048 Hz. EEG was recorded at least once during one of the following three experimental conditions: at rest, sensory stimulation (somatosensory, auditory or photic) and performance on cognitive tasks (Table 3). Some studies reported the use of more than one experimental condition for EEG recording.

qEEG findings.

The most frequently analyzed parameters of brain activity were EEG power spectra at rest and the magnitude of different components of the event-related potentials (ERP) elicited by sensory stimulation or cognitive tasks. A brief description of the findings obtained at rest and in response to external stimulation is listed below. Nevertheless, it should be noted that the heterogeneity of the outcome variables prevented the standardization of data, making it impossible to conduct a meta-analysis.

Regarding spontaneous brain activity, eight studies collected spontaneous EEG data by asking participants to rest with either opened or closed eyes. The main EEG parameters analyzed in those studies were power spectra (or power density) [2,8,16,20,23,25] (n = 6) and peak frequency (n = 5) [2,9,10,20,23]. The analyses of power spectra revealed that chronic pain patients displayed increased theta (3.75–9 Hz) [8,16,20,25], alpha (7.75-13hz) [20,23,25], beta (12–30 Hz) [8,20], or delta band power (0,5–4 Hz) at rest [20].

Five studies were specifically designed to assess differences on spontaneous EEG by computing the dominant peak frequency from the average power spectra [2,9,10,20,23]. The major relevant finding from these studies was that the dominant peak of the power spectra shifted towards lower frequencies in chronic pain patients as compared with healthy controls.

Moreover, most studies on power spectra of spontaneous EEG activity have demonstrated that chronic pain patients displayed enhanced power spectra at frontal and parieto-occipital electrode locations. One study further compared the differences between chronic pain patients and controls on global spectral power by computing an estimation of the neural generators (source localization) within low theta (4–6 Hz), high theta (6–9 Hz), alpha (9–12 Hz), low beta (12–16 Hz), and high beta frequency band (16–30 Hz) [8]. The authors found that over-activations within high theta and low beta frequency bands were localized to multiple pain-associated areas, including insula, anterior cingulate, prefrontal, inferior posterior parietal, primary, secondary, and supplementary somatosensory cortices.

Some of the included studies assessed brain activity elicited by sensory stimulation or cognitive tasks. We included in this subsection all the studies using event-related potentials (ERPs) as potential markers of brain processing in chronic pain. Eight studies reported ERP changes when comparing patients with chronic pain and healthy controls [15,17–19,21,22,24,25]. Four studies reported significant reductions of several ERP components in response to different sensory and cognitive tasks [15,17,21,25]. These studies used ERPs to answer specific questions about sensory, affective and cognitive processing, making comparisons among them difficult.

The experimental tasks used in ERP studies included: photic stimulation during specific periods of migraine attacks [15,18]; auditory and somatosensory stimulation in individuals with fibromyalgia [19]; performance on cognitive tasks with varying difficulty in individuals suffering from different types of chronic pain [24]; reading pain descriptors in patients with fibromyalgia and musculoskeletal pain [22]; looking at faces expressing happy/anger/pain in individuals with fibromyalgia [17]; viewing pleasant/unpleasant/neutral pictures in patients with musculoskeletal pain [21]; and imagining painful/non-painful movements of limbs in participants with fibromyalgia [25].

The outcome variables used in these studies were also varied and included several parameters of the brain activity such as spectral [9,15,25]; amplitudes of early (P50) and late components (N100, P2, P300) of the ERPs (17,19,21,22,24]; and temporal dynamics of EEG signals—coherence [18], and non-linear parameters of brain activity such as fractal dimension and entropy [21].

Changes in EEG activity occurred in several brain areas and in more than one region in most studies. Changes elicited by photostimulation were found in O1 and O2 electrodes [15], as well as in P3, P4, and F4 electrodes [18]. Changes in brain activity elicited by cognitive were observed in almost all electrodes: Fz, Cz, Pz, and Oz [24]; T7, T8, P7, P8 [17], CZ [25], and F4, C4, CP4, TP8, P4, O2, F3, C3, CP3, TP7, P3, O1 [22]. Changes elicited by sensory stimulation were found at Fz, F3, F4, Cz, C3, C4, Pz, P3, P4 [19] and C3, C4, P3, P4, F3 F4 electrodes [21]. As changes of brain activity elicited by a stimulus or a task are very specific to each study, detailed information is provided in the description of each article in the next section (General overview of included papers).

Several of the studies computed correlations between behavioral or psychological features of pain and brain activity. For instance, significant positive correlations were found between pain intensity, delta power and asymmetry in migraine patients [16]. EEG power was negatively correlated with trigger sensitivity score, photophobia and family history of migraine [15]. In these patients, the asymmetrical distribution of enhanced power to photic stimulation was also negatively correlated with pain intensity. Disease duration was negatively correlated with the shift of the alpha dominant peak in chronic pancreatitis patients [2], but positively correlated with EEG power in back pain patients [10]. Frequency indices (frequency of the dominant peak and center of gravity) were negatively correlated with psychopathological symptoms and pain perception, and positively correlated with life satisfaction [10].

Table 3 describes all results for each study. S2 Table contains additional information, including a description of covariates.

General overview of included papers.

A brief description of major findings reported by the selected studies is provided below. The first group of studies assessed spontaneous brain activity through EEG recordings.

Sarnthein et al. (2006) studied 15 patients with chronic neurogenic pain and 15 healthy controls (HC). Their experimental protocol included the evaluation of EEG power spectra at rest with opened (EO) and closed eyes (EC). They showed that patients had increased average power density in delta, theta, alpha, and beta ranges (from 2-25Hz). The peak frequency was shifted towards lower values in the presence of neurogenic pain. Discriminant analysis of peak frequency and peak height successfully identified patients, but it was not possible to use analysis of variance, due to non-parametric data distribution. These findings might also be related to the use of central acting drugs, like antiepileptics and benzodiazepines. Alpha peak was negative for HC in the parietal and occipital regions, and higher in the 13-15Hz range (high alpha/low beta) in frontal electrodes. Theta-beta coupling was abnormally high at fronto-parietal electrodes in patients, which was reversed by central lateral thalamotomy. This surgery successfully decreased pain in most of the studied patients (not all sample participants were submitted to surgery), decreasing theta power. However, it is possible that this result might have occurred because of the withdrawal of antiepileptic drugs. The authors proposed that the main findings on theta peak frequency and peak height was due to increased thalamocortical dysrhythmia.

Studying the same sample of neurogenic pain as Sarnthein et al. (2006), Stern et al. (2006) investigated 16 patients pre and post central lateral thalamotomy, compared to 16 HC. Data were analyzed by tomographic LORETA in the theta and beta ranges, in 15 regions of interest related to pain processing (pain matrix). They founda higher peak activation in the periinsular parieto-temporal cortex in low theta range (4-6Hz) for patients. The same occurred at low-beta range in the anterior cingular cortex, dorsolateral prefrontal cortex and left insula. Frequency dependent overactivation was shown in the 7-11Hz range (alpha-theta) for all regions, followed by a less intense peak in the lower-beta range (12-16Hz) in both right and left anterior cingulate, right dorsolateral prefrontal, mid cingulate cortices and posterior insula. Primary and secondary sensory cortices showed stronger activation in the left side. The use of central acting medications did not substantially change the results. Therapeutic central lateral thalamotomy lead to decreased overactivation of the right anterior cingulate cortex (8Hz–theta) and mid cingulate cortex (14Hz–beta).

Boord et al (2008) assessed participants with paraplegia, with (n = 8) and without (n = 8) neuropathic pain, comparing their resting EEG activity with healthy controls (n = 16). They found that peak theta-alpha frequency was shifted to lower frequencies at all recording sites in the presence of neuropathic pain, proposing that this phenomenon represented the presence of thalamocortical dysrhythmia. The use of medication was related to a decrease in theta-alpha frequency at three electrode locations (Pz, P3 and P7). They also analyzed eyes opened/eyes closed (EO/EC) ratio at all electrode locations in a broadband range (1-40Hz). Reactivity was decreased in the participants with neuropathic pain in 11 recording sites when compared to healthy controls, and in central (C3, C4) and frontal (F3, Fp1) electrodes, when pain participants were compared to participants without pain. These changes occurred in delta, theta, alpha and beta frequency ranges. The authors conclude that the slowing of EEG frequencies in neuropathic pain participants was attributed in part to the use of medications, but also as a consequence of pain. They also propose that EO/EC reactivity is a useful indicator of thalamocortical function, suggesting that there is a failed mechanism in the neurophysiological adjustment to sensory inputs in the neuropathic pain condition.

Bjork et al. (2009) compared interictal EEG recording from 33 migraineurs (6 with and 27 without aura) and 31 controls to estimate delta, theta, and alpha spectral power outside the period of crisis. They also evaluated absolute inter-hemispheric asymmetry. Data was collected at rest with closed eyes. Although they did not identify the number of electrodes utilized, they were grouped in three areas (parieto-occiptal, fronto-central and temporal). Differences between the groups were found only regarding increased theta power (all three regions), which was more pronounced in migraneurs without aura. Other secondary findings included a positive association between headache intensity, delta power, and delta asymmetry. An inverse correlation was also found between headache history, age and delta power/asymmetry. They argue that increased theta power may represent decreases in the frequency of thalamic activity, and delta findings represent latent cortical spreading depression.

Mendonça-de-Souza et al., (2012) assessed 11 women with migraine and visual aura, and seven HC. Their methods included EC resting EEG evaluation. They assessed partial coherence among occipital, parietal e frontal electrodes, within the frequency ranges of delta, theta, alpha, beta and gamma. Their results showed that in the basal period, migraneurs presented increased coherence in fronto-parietal and fronto-occipital networks. If the direction of the coherence was not considered, this phenomenon occurred in all frequency ranges, from delta to gamma.

Schmidt et al. (2012) assessed 37 low back pain participants and 37 HC to investigate the presence of thalamocortical dysrhythmia in the pain group. Their experimental protocol included 60 EEG electrodes recording at rest, with EO and EC. The clinical outcome variables were pain, quality of life and life satisfaction. EEG variables were average peak frequency, peak alpha power, peak alpha frequency, and overall power. They did not find significant differences in EEG indices between patients and controls, even after stratifying the patients into subgroups with and without neuropathic, or by pain severity. Correlations were only significant between EEG power and pain intensity and the presence of psychopathology. They conclude that thalamocortical dysrhythmia is not generally found in patients with moderate low back pain.

De Vries et al. (2013) compared the resting state EEG data from 16 patients with persistent abdominal pain due to chronic pancreatitis and 16 matched HC. They used 26 electrodes to acquire EEG data with EO and EC. Electrodes were grouped in central, parietal, occipital and frontal regions of interest and peak alpha frequency was estimated based on the center of gravity method. They did not find differences in alpha power between groups, but only in alpha peak frequency, which was skewed to lower frequencies in patients, predominantly in the parieto-occipital regions. Pain duration was inversely correlated with alpha peak frequency. They conclude that peak alpha frequency measures may represent a biomarker of chronic pain.

van den Broeke et al. (2013) studied two groups of patients after mastectomy, eight with pain and 11 without pain. They recorded EEG with 64 channels at rest and again, after a painful stimulation. This review only considered baseline data. Overall alpha EEG activity and alpha peak frequency (computed by the center of gravity method) were recorded only for parieto-occipital electrodes. Overall alpha amplitude, but not alpha peak frequency was higher in patients with pain. Neither alpha amplitude or frequency were correlated with pain intensity. However, pain intensity ranged from 3-6/10, which is considered to be low to moderate and is different from other studies where patients had severe neuropathic pain. The authors stated that their results are not sufficient to establish a clear role of alpha changes as a pain biomarker.

Vuckovic et al. (2014) evaluated two groups of 10 paraplegic patients, one with central neuopathic pain and another without. A third control group without any neurologic injury or pain was also enrolled. They recorded spontaneous EEG activity with 61 electrodes in EO state. During this state participants visually focused on a small cross. Paraplegics with pain showed larger theta power than those without pain, but were not different from controls. Patients with pain also presented greater alpha power than patients without pain. The dominant alpha frequency was lower in paraplegics with pain compared to those without pain. The authors suggest a specific EEG pattern for chronic pain.

A second group of studies assessed brain activity related to task performance elicited by sensory stimulation or cognitive tasks. They are described in the next paragraphs.

Montoya et al. (2006) examined brain habituation to repetitive, non-painful tactile and auditory stimulation in a sample of 15 female patients with fibromyalgia (FM) and 15 matched HC by using evoked-related potentials (ERP). Results indicate that amplitudes of early (P50) and middle somatosensory ERP components (160–360 ms) were reduced after repetition in HC, but not in FM patients. In addition, ERP amplitudes to auditory stimuli were reduced after repetition in both groups. The authors suggest the existence of an abnormal brain mechanism to specifically inhibit irrelevant somatosensory information in FM.

Veldzhijzen et al. (2006) studied 14 patients with either neuropathic or nociceptive chronic pain and 30 HC. They investigated whether resources are shared by pain and attention, and if allocation resources during pain are deficient. They studied ERP elicited by easy and difficult tasks associated with the recognition of pictures. Their results showed that HC presented smaller P300 amplitudes in Oz electrodes in the difficult task, and pain patients presented the reverse response. P100 amplitudes were larger in HC at Pz. The authors conclude that pain patients have abnormal attention allocation, but not attention capacity.

Sitges et al. (2007) examined the effects of negative mood on the brain processing of affective and sensory pain-related descriptors in 18 female patients with FM, 18 with chronic musculoskeletal (MSK) pain (rheumatoid arthritis, radiculopathy, herniated disk) and 16 female HC. Visual ERP elicited by pain descriptors and pleasant words were recorded while participants were making a decision about their appropriateness to define their pain experience (self-endorsement task). Results indicated that chronic pain patients reacted slower than HC participants, and that pain descriptors elicited higher positive ERP amplitudes than pleasant words in HC and MSK patients, with no differences in FM patients. These findings suggest an altered involvement of attentional and motivational brain systems in FM patients, as well as an excessive engagement of the motivational system during the processing of affective salient stimuli in patients with MSK pain.

Sitges et al. (2010) analyzed the brain dynamic of affective modulation of somatosensory processing in 19 female chronic pain patients (MSK) and 21 female HC. The experimental protocol included the recording of somatosensory ERPs elicited when participants were viewing affective pictures (pleasant, unpleasant and neutral). Results showed that pleasant pictures elicited an enhancement of the somatosensory P50 amplitudes in HC but not in patients with MSK pain. Moreover, power spectra of delta, alpha-2, beta-1 and beta-2 EEG frequencies were overall higher in HC compared to MSK patients. Finally, entropy and fractal dimension of EEG signal displayed a more asymmetrical distribution (right > left hemisphere) in MSK patients than in HC. These findings suggest that brain responses to somatosensory stimulation in chronic pain patients are abnormally modulated by the affective characteristics of the context in which body stimulation occurs. The reduction of alpha, beta and delta power, as well as the increased entropy over the right parietal lobe in MSK patients might indicate that persistent pain would lead to a sustained abnormal activation of the attentional brain system during the processing of non-painful body information.

Bjork et al. (2011) studied steady state visual evoked EEG-responses for 6, 12, 18 and 24 Hz flash stimuli in 33 migraineurs without aura, eight migraineurs with aura and 32 HC. EEG was recorded with 21 electrodes in an EC state. Intermittent photic stimulation was delivered for 10 seconds at each frequency, and repeated two times. Driving power (EEG power during photic stimulation) was analyzed and correlated with sensory hypersensitivity and severity of migraine. Driving responses to 18Hz and 24Hz were suppressed in the interictal phase in migraneurs without aura when compared to controls, probably representing a low-activation of the cerebral cortex. The same occurred at 18Hz when migraneurs were compared to controls, irrespective to the presence or not of aura. Additionally, the 18Hz driving power was more symmetric in migraneurs without aura when compared to those with aura or controls. Those participants with low threshold to photic stimulus also presented symmetric hemispheric activity, and were prone to present high sensitivity to light, sound, smell and physical activity. All these findings suggest a relationship between deficient habituation in migraneur patients.

Mendonça-de-Souza et al., (2012) also evaluated EEG inter-hemispheric coherence after 9Hz photic stimulation for 11 women with migraine and visual aura compared to seven HC. Photic stimulation was held through four rounds of 9Hz stimulation (3 seconds each, 0.2J, 20 cm in front of the eyes). EEG was recorded before, during and after stimulation, but only the latter was used for analysis. They showed that coherence decreases when photic stimulation starts in migraneurs, the opposite occurring in HC. When the stimulation ceased, migraneurs participants showed higher coherence than controls. They interpret these results as a resilient brain mechanism to deal with photic stimulation, a potential crisis triggering stimulus for migraine patients.

Gonzalez-Roldan et al. (2013) analyzed brain responses to viewing facial expressions of others’ pain in 20 female fibromyalgia patients (FM) and 20 HC by using visual ERPs. FM patients displayed higher N100 amplitudes to pain and anger faces as compared to neutral faces, as well as more reduced N100 amplitudes to happy faces in comparison with HC. By contrast, ERP amplitudes in the time-window 200–300 ms were more positive in response to happy than to the other faces in HC, but no differences were observable in FM patients. In addition, FM patients displayed higher theta power in response to pain and anger faces, as well as reduced alpha to all faces in comparison with HC. According to the authors, these findings suggest that FM patients could be characterized by an increased mobilization of early and preconscious attentional resources to biologically relevant cues during encoding processes and retrieval of sensory information

Vuckovic et al. (2014) also recorded EEG activity in EC and EO state (at rest) for three groups, including 10 paraplegic patients with central neuropathic pain, 10 paraplegic patients without central neuropathic pain, and 10 participants without any neurologic injury or pain. In the EC state, participants had to imagine hand or lower limb movements. There were no differences among the three groups regarding EEG power at this EC state. Paraplegic patients with or without pain showed no differences in alpha power in the EC state (imagery), compared to the EO state (rest), at posterior electrodes, which should increase in this state. This finding highlights a deficiency in EC/EO ratio. The dominant alpha frequency was lower in paraplegics with pain compared to those without pain. Event related desynchronization after imagination of movements was diffuse in the patients with pain, with no event related synchronization in the surrounding areas. The authors suggest this as a specific EEG pattern for chronic pain.