Association of widespread pain and peak alpha frequency cannot be isolated to a single network.

Peak alpha frequency (PAF), a measure derived through electroencephalography (EEG), has garnered increased attention as a pain-related measure of cortical activity. 6 – 8 , 12 , 31 Multiple methods exist for calculating PAF, which generally fall into 2 categories: channel-based and component-based. Channel-based methods include the global average (averaging PAF across all electrodes) and region of interest (ROI) (averaging PAF across predefined scalp regions), whereas component-based methods extract PAF from independent components identified through independent component analysis (ICA). 14 , 32 Although channel-based methods have shown associations between PAF and pain sensitivity in healthy populations, recent work has increasingly favored the sensorimotor component. 7 , 11 , 14 This method, which uses ICA to isolate sensorimotor-related activity into a component, has demonstrated strong correlations with pain sensitivity in recent experimental pain models and has been used to predict individual sensitivity in healthy participants. 7 , 11 , 14 For instance, 1 study compared 2 variations of the sensorimotor component with a channel-based sensorimotor ROI and found that the sensorimotor component approach more accurately predicted pain sensitivity. 7 These findings suggest that pain-relevant cortical activity may be localized within the sensorimotor network in healthy individuals. In contrast, the relationship between PAF and chronic pain is not as well established. Although several studies have compared between healthy and chronic pain populations, findings have been inconsistent with some showing increased or decreased PAF and others showing no difference. 32 Recent studies have shifted focus to differences in subtypes of chronic pain groups, examining diagnoses and distribution of pain. 6 , 31 For example, Cavaleri et al. (2025) found individuals with chronic pelvic pain had significantly reduced global average PAF if they had widespread pain compared with those with localized pain. Notably, this reduction was observed across all EEG electrodes with the greatest differences in the sensorimotor and occipital regions. 6 These findings suggest that PAF may be more closely linked to the spatial distribution of pain, particularly widespread pain rather than to the presence of chronic pain itself and could help explain contradictory findings in previous studies. Studies have also reported associations between widespread pain and greater pain sensitivity in chronic pain groups. 10 , 26 Although not yet tested in chronic pain populations, the strong relationship between sensorimotor component PAF and pain sensitivity in healthy populations raises the possibility that it may be effective in distinguishing between widespread and localized pain, potentially even more so than the global average. Although multiple analytic and acquisition choices can influence PAF estimates, prior work indicates that the factors most relevant for interpreting pain-related PAF differences are frequency estimation methods (center of gravity [COG] vs peak picking) and spatial methods (global average vs ROI) used to derive the signal. Since COG provides more stable and reliable estimates than peak picking and is the dominant approach in pain-related PAF studies, this study focused specifically on the spatial variation of PAF derivation. 6 – 8 , 11 , 14 , 31 , 32 To the best of our knowledge, there has been no comprehensive comparison of spatial variations of PAF in chronic pain populations to determine which approach best explains pain-related outcomes. To address this gap, the present study performs a secondary analysis building off a previous study using a preexisting dataset to compare multiple spatial variations of PAF and their relationship with widespread pain in a visceral chronic pelvic pain group. We hypothesize that although both global average and sensorimotor component PAF will differentiate widespread from localized pain, the sensorimotor component will best explain these differences, given its established link to pain sensitivity and prior evidence that individuals with widespread pain exhibit heightened sensory sensitivity—including to visual, auditory, and tactile stimuli. 7 , 17 , 27 , 33 , 37

The dataset used in this study has been previously published in Cavaleri et al. (2025). The present work constitutes a secondary analysis of this dataset, focusing on comparing variations of spatial PAF derivation across pain groups. Participant recruitment, data collection procedures, and initial findings are described in detail in the original report. 6 Briefly, 38 participants (19 male and 19 female) with urological chronic pelvic pain syndromes (UCPPS), a visceral pelvic pain condition, met eligibility criteria. Eligibility criteria followed those established by the Multidisciplinary Approach to the Study of Chronic Pelvic Pain Research Network and are fully described in Cavaleri et al. (2025). Key criteria include participant age ranges from 18 to 65 years old with a clinical diagnosis of UCPPS. Female participants were required to have symptoms present most of the time for the past 3 months, whereas male participants had to have symptoms present most of the time in any 3 of the past 6 months. Individuals were excluded from the study for any neurological disorders affecting bladder or bowel function, autoimmune or infectious disorders, active psychiatric disorders, severe cardiac, pulmonary, renal or hepatic diseases, a history of pelvic radiation, bladder cancer, or infectious cystitis. Women with gynecologic pain conditions unique to females (eg, endometriosis) were excluded from this study to allow sex-balanced matching. The participant sample was derived from 2 independent datasets collected at different times using similar experimental protocols. Only resting-state, eyes-closed EEG and corresponding demographic and pain-related survey data common to both cohorts were used in the present study. Differences between data collection between the 2 cohorts are further highlighted in the earlier work. 6 Males and females were matched for age and degree of widespread pain with a propensity matching score, which is further detailed in the cited publication. Participants were recruited through community advertisements and contact with physical therapists and urologists for referrals throughout the Los Angeles, California area. All participants were provided with written and verbal descriptions of the experiment, and written informed consent was obtained before testing. All procedures were approved by the USC Institutional Review Board and performed in accordance with the Declaration of Helsinki. Data were obtained during a single experimental session including demographic information, self-reported pain intensity, and EEG recordings. Participants completed self-reported body maps to assess the spatial distribution of their pain. Demographic information, self-reported pain intensity, and self-reported body maps were collected before EEG recordings. The experimental protocol is further detailed within the previously cited publication. 6 Participants reported current pain intensity using a previously validated 11-point numerical rating scale in which 0 = no pain and 10 = worst pain imaginable. Pain duration was measured by asking participants when the onset of their pelvic pain was and taking the difference between this and the day of collection. As stated in our previous publication, pain distribution was determined through participant completion of body maps with the Chronic Overlapping Pain Conditions Screener (COPC-S) and the Brief Pain Inventory (BPI). 6 To ensure consistency between the 2 measures, the BPI body maps were transposed onto the COPC-S body map. Further details can be found in the Supplemental Files ( http://links.lww.com/PR9/A399 ) of Cavaleri et al. (2025). Based on previous research, individuals with greater than 2 regions of pain were put into the widespread pain group, whereas those with 2 or less regions of pain were put into the localized pain group. 25 , 34 Electroencephalography data were collected using a 64-channel, ANT Neuro gel-based electrode cap with sintered Ag/AgCl electrodes. Data from both populations were collected in a supine resting position with eyes closed for at least 5 minutes. Electroencephalography was collected with a sampling rate of 2048 Hz and acquired with eego sports acquisition software (v1.2.1) from an Ant Neuro eego Sports amplifier (Product Number ee-202) as previously described. 6 All EEG data were reprocessed from the raw data. Electroencephalography preprocessing followed the same pipeline used in the prior publication, which selected parameters from a comprehensive review of PAF analyses in pain studies. 6 Since ICA and Multiple Artifact Rejection Algorithm involve stochastic processes, small differences in PAF values relative to the previously published report are expected and fall within normal analytic variability. 32 Data processing was executed with custom scripts in EEGLAB (v2024.0) and MATLAB (2024a). Data processing was conducted across all 64 electrodes for each participant. The pipeline was as follows: bandpass filter (1–100 Hz), notch filter (58–62 Hz), ICA, the Multiple Artifact Rejection Algorithm, and finally signal was re-referenced to the common average of all channels for each participant. 6 Across all spatial variations, PAF was determined using the restingIAF MATLAB function using a 2-second Hamming window with a 1-second overlap. The alpha band was set at 7 to 13 Hz, and the COG output was used as the measure of PAF as previous work has shown this measure of PAF to be the most stable. 4 , 8 , 24 The COG is determined as the weighted sum of the alpha spectrum divided by the total power and has been widely used across EEG pain literature. 32 Peak alpha frequency was determined through 4 different types of spatial variations: global channel average or grand average, ROI, sensorimotor component, and group independent component analysis (GICA). The global channel average takes an average of PAF across all 64 electrodes for each participant. This method was used in the previous publication identifying a significant difference in PAF between the widespread and localized pain groups. 6 , 31 This method has also been widely used to measure pain-related outcomes and has been shown to be a highly stable method of deriving PAF. 6 , 31 , 32 Similar to the global channel average, ROI PAF was determined by averaging PAF across a select group of electrodes for each participant. Within this analysis, the frontal (electrodes: F3, Fz, F4), sensorimotor (electrodes: C3, Cz, C4), parietal (electrodes: P3, Pz, P4), and occipital (electrodes: O1, Oz, O2) regions were identified, in line with previous pain studies. 32 The sensorimotor ROI, specifically, has been used to identify differences in pain sensitivity within healthy populations as well as differences in PAF between chronic pain groups. 14 , 32 We determined the sensorimotor component PAF using a custom MATLAB template-matching algorithm that matched ICA scalp topographies to an EEG.icawinv matrix created through participant data in EEGLAB. In this analysis, we used a sensorimotor spatial template similar to the sensorimotor template used in Chowdhury et al. (2025) (Fig. 1 A). The sensorimotor template scalp topography was normalized to unit length and compared with each participant's normalized component scalp topographies using cosine similarity, quantified as the dot product of unit vectors, to determine spatial similarity scores on a scale from 0 to 1. The component with the highest similarity (closest to 1) was selected as the sensorimotor component for the participant. Components with lower similarity scores were manually inspected by evaluating their topography, spectral profile, and time course. In 1 participant, no plausible sensorimotor component was identified and was excluded from the sensorimotor component PAF analysis but retained in all others. Peak alpha frequency was then extracted for each of these components. The sensorimotor component has been identified most commonly through visual inspection of component head maps for each participant. 7 , 13 However, more recently, 1 study used an automated component selection method that chose the component with the strongest spatial correlation to a sensorimotor template originating from a prior work. 7 Comparison of global channel average and sensorimotor component PAF. (A) Example of determining a sensorimotor component within each individual. The sensorimotor template was used within a template matching algorithm to identify the head map that best matched the template topography. The highlighted component indicates the head map chosen by the algorithm as the sensorimotor component for that participant. (B) Bars were created using average PAF in widespread and localized pain groups as calculated using global channel and sensorimotor component spatial variations of PAF. A linear regression model was run to determine how the degree of widespread pain affected PAF while controlling for age and sex. There is a significant difference in average PAF between pain groups using the global average method, but not when using the sensorimotor component. Finally, we used group ICA as a novel measure within PAF-pain literature to explore methods of PAF entirely and to determine any potential, underlying neural patterns across the participant group. The GICA is primarily used within functional magnetic resonance imaging literature to determine whole group neural patterns and can be powerful in identifying trends that are not obvious through other methods of analysis. EEGIFT was used to process and run the GICA and provide the outputs. Similar to the previous methods, PAF was calculated with custom MATLAB scripts from the outputs given by EEGIFT. Since GICA requires all participant data to be the same length, EEG data were first trimmed to an equal time series length using a custom MATLAB script. EEGIFT parameters can be found within the Supplemental digital content ( http://links.lww.com/PR9/A399 ). GICA requires the number of components created to be input by the user. This inherently introduces some error as it is unknown what the optimal number of components should be, and it forces data to be configured into this exact number of components. Therefore, multiple iterations of the GICA were performed with 5, 10, 15, 20, and 30 ICs created across participants. Peak alpha frequency was extracted from participant time series data generated by the GICA analysis. Since GICA does not guarantee that a sensorimotor-related component will emerge at a specific component index, all components generated at each dimensionality (5, 10, 15, 20, and 30 ICs) were retained for statistical testing. Each component PAF was entered individually into the regression model to assess whether any group-level component differentiated widespread and localized pain. Statistical analyses were performed using custom MATLAB and RStudio scripts. Group differences in age, current pain intensity, and pain duration were reported previously in Cavaleri et al. (2025) using independent-samples t-tests; however, since the present study uses the same participant sample, these values were carried over without reanalysis. For each spatial variation, a linear regression model was performed to determine differences in average PAF between pain distribution groups, while controlling for age and sex. For GICA-based methods, P -values for each component were corrected for multiple comparisons using the Bonferroni method. Pain intensity and pain duration outcomes were added to regression models to understand the influence of these variables on PAF outcomes. Cohen d effect size was calculated for each spatial variation to identify the PAF methods that best explained differences in average PAF between widespread and localized pain groups. For all group comparisons, 95% confidence intervals and corresponding test statistics were extracted from the fitted linear regression models or from independent-samples tests when appropriate.

Participant demographic information is found in Table 1 . As discussed in our previous work, the sample was composed of 38 participants from 2 independent datasets split into widespread (n = 24) and localized (n = 14) pain groups based on COPC-S and BPI body maps. 6 Each group contains an equal proportion of males and females (widespread: 12 males, 12 females; localized: 7 males, 7 females) and were matched based on age and pain distribution. Pain groups did not significantly differ in age, pain duration, or current pain intensity. Participant demographics. Participants were divided into widespread and localized pain groups based on the number of areas participants had experienced prolonged pain on body maps as discussed in Cavaleri et al. (2025). There were 4 participant entries missing for Pain Duration (3 from widespread and 1 from localized) totaling 21 participants in the widespread group and 13 in the localized group for this outcome measure only. Test statistics are reported for age, pain duration, and current pain intensity comparisons. As this is a secondary analysis of previously collected data, the results below focus specifically on comparisons across spatial variations of PAF. Figure 1 B shows that, after controlling for age and sex, global channel average PAF was significantly lower in the widespread compared with the localized pain group (mean ± SD: widespread = 9.56 ± 0.7 Hz, localized = 10.15 ± 0.6 Hz; P = 0.012). Analyses of the ROI average PAF (frontal, sensorimotor, parietal, and occipital) also showed significantly lower PAF in the widespread vs the localized pain group after controlling for age and sex (Fig. 2 A). Mean and standard deviation values for widespread and localized pain group ROIs were as follows: frontal (widespread = 9.52 ± 0.7, localized = 10.1 ± 0.58 Hz, P = 0.012), sensorimotor (widespread = 9.54 ± 0.69 Hz, localized = 10.08 ± 0.58 Hz, P = 0.019), parietal (widespread = 9.67 ± 0.72 Hz, localized = 10.28 ± 0.66 Hz, P = 0.016), and occipital (widespread = 9.62 ± 0.70 Hz, localized = 10.24 ± 0.66 Hz, P = 0.01). Cohen d effect sizes were calculated for all 5 channel-based PAF methods (Fig. 2 B) ranging from 0.83 to 0.91, which are all considered high. 9 Given the unequal group sizes (n = 24 widespread; n = 14 localized), the sample had 80% power to detect effects of approximately |d| ≥ 0.75 to 0.8; the observed channel-based effects were larger (d = 0.83–0.91; CI range across channel-based methods: 0.09–1.09; Table 2 ). All confidence intervals remained positive, indicating a consistently higher PAF in the localized group. Although lower bounds extended into the small-effect range, the observed effects exceeded the detectable range for this sample. Comparison across electrode regions of interest. (A) Comparison of average PAF across select ROIs showing significant differences in PAF between localized and widespread groups while controlling for age and sex. (B) Head map representing location of electrodes on head with electrodes included in ROIs corresponding with colored boxes. Calculated Cohen d effect size for each method. Effect sizes are high and similar to each other indicating across regions of the scalp, pain distribution is well explained by these methods, but there is little difference between them. Statistics across spatial variations of PAF. PAF for ROI methods was averaged across electrodes within ROI before averaging across participants within respective groups. PAF for GICA methods was selected from the component with the lowest P-value after correcting for multiple comparisons. If all components for an iteration of GICA had a P-value of 1 after correction, the component with the lowest P-value before correction was selected. Average PAF and standard deviation were calculated from groups based on this component. The “Widespread” and “Localized” columns report mean PAF values (Hz) for each method, with standard deviations in parentheses. Test statistics and 95% confidence intervals are reported for each comparison. Bolded effect sizes indicate those that are high and explain pain distribution across individuals with chronic pain. Significant difference between widespread and localized PAF. Minimum Bonferroni-corrected P-value across components. Compared with channel-based methods, the sensorimotor component showed only a minimal reduction in average PAF in the widespread group (9.96 ± 0.57 Hz) relative to the localized group (10.13 ± 0.52 Hz), and this difference was not statistically significant ( P = 0.35). The corresponding effect size was small and nonsignificant (Cohen d = 0.31, 95% CI [−0.19, 0.54]; Table 2 ), consistent with the null result from the regression model. Similarity scores ranged from 0.36 to 0.92 (mean = 0.76, SD = 0.13). Although 4 participants had moderate similarity scores (range: 0.36–0.54), manual inspection indicated that 3 of these components still exhibited plausible sensorimotor topographies; the 1 participant component that did not was excluded from the sensorimotor component analyses. To determine whether broader component-level decompositions would capture group differences, we next evaluated PAF derived from GICA. For each dimensionality, the component with the smallest P -value is reported in Table 2 as a summary indicator, whereas Figure 3 displays the full distribution of effect sizes across all components. Across all component dimensionalities, GICA-based PAF failed to differentiate between widespread and localized pain groups (Table 2 ). Figure 3 shows that, across multiple iterations of GICA with varying numbers of independent components, no component exhibits an effect size comparable with those from channel-based PAF measures. The GICA-derived component with the largest absolute effect size showed a modest effect (d = −0.49, 95% CI [−0.76, 0.13]; Table 2 ). Interestingly, GICA analyses showed a reversal in trend, with the widespread group having a higher average PAF compared with the localized group (Table 2 ). This is reflected in the effect sizes for this method as they are all negative while channel-based and sensorimotor component methods have positive effect sizes. GICA components Cohen d plot. (A) Examples of component head maps across participants for the 10- and 20-component GICA. (B) Plotted Cohen d effect size for each component across multiple iterations of a GICA analysis with varying numbers of independent components (5–30 ICs). Linear regression models and Cohen d effect sizes were calculated for each component controlling for age and sex. None of the components from any of the iterations were similar to that of the global channel Cohen d marked by the red dashed line. The prior publication investigated the influence of pain intensity and duration on the global channel average PAF in which there was no significant P -value. 6 This analysis was repeated for the other spatial variations of PAF already discussed: ROI, sensorimotor component, and GICA. Similar to the earlier analysis, 6 neither pain intensity nor pain duration had a significant relationship with any of the spatial variations of PAF (see supplemental digital content, Table 6, http://links.lww.com/PR9/A399 ). This indicates that these 2 variables are not related to PAF and pain distribution within this population.

This study is the first to compare multiple spatial variations of PAF in individuals with widespread and localized UCPPS. Consistent with prior work, 6 the global average PAF was significantly lower in widespread pain, but contrary to our hypothesis, the sensorimotor component PAF was not significantly different between groups. Importantly, this pattern was not limited to the global average as each ROI channel-based analysis also showed similar differences in PAF to the global average while GICA-derived PAF did not. Together, these findings suggest that the extent of pain distribution could be related to the degree of altered cortical alpha rhythm in an individual, which is not fully captured by the sensorimotor component alone. In addition, there was no effect when adding pain-related outcomes into the linear regression model. These findings provide novel insight into the relationship between cortical alpha oscillations and chronic pain, specifically in individuals with widespread pain. Based on prior literature, we hypothesized that the sensorimotor component PAF would differentiate between individuals with localized vs widespread pain. 7 , 13 , 14 However, our analyses did not reveal a significant group difference. Sensorimotor component PAF has been linked with pain sensitivity in healthy populations across multiple quantitative sensory testing (QST) paradigms. 7 , 13 , 14 Although QST methods have shown significant differences in pain sensitivity between chronic pain and healthy populations, 19 , 38 – 40 it is not well demonstrated that similar differences exist between chronic pain groups with widespread and localized pain. Furthermore, although some studies report greater pain sensitivity in individuals with widespread pain, these findings are often confounded by sample size, psychosocial factors, and variability across multiple QST modalities. 15 , 17 , 26 In addition, evidence from a study pulling from the TwinsUK database demonstrated no significant differences while using multiple QST modalities similar to those used in the previously mentioned studies, further emphasizing the inconsistency of findings for this relationship. 35 This lack of consistent evidence may help explain why sensorimotor component PAF did not differentiate between pain groups in the present study. Notably, sensorimotor component PAF has not yet been evaluated as a marker distinguishing chronic pain and healthy control groups, and exploring this comparison may help clarify its role in pain sensitivity and underlying neural mechanisms. Despite the lack of group differences in sensorimotor component PAF, the sensorimotor ROI showed significant differences, indicating the spatial approach to derive PAF substantially influences sensitivity to chronic pain–related cortical activity. Structural evidence supports this distinction with observed localized changes in gray matter volume within sensorimotor subregions between individuals with widespread and localized pain, highlighting the presence of region-specific alterations between chronic pain groups. 25 However, ICA-derived components are designed to be statistically independent, meaning that shared or overlapping activity across regions is intentionally minimized. 20 , 29 As a result, pain-related neural alterations that are spatially distributed, heterogeneous, or weakly expressed across multiple generators may not be fully captured within a single independent component. 5 The sensorimotor cortex contains functionally distinct regions with varying patterns of connectivity, 16 , 18 making it difficult for a single component to reliably capture this heterogeneity. In contrast, channel-based approaches such as the sensorimotor ROI aggregate overlapping activity across the region, which may enhance sensitivity to group-level differences and distributed cortical alterations. 1 , 2 This framework may explain why GICA-derived PAF failed to differentiate pain groups across component dimensionalities. Overall, component-based PAF methods may be less sensitive to localized and distributed cortical changes associated with widespread chronic pain. Beyond the sensorimotor cortex, individuals with widespread pain have also demonstrated altered activity and functional connectivity across neural networks, including salience, default mode, as well as regions such as the insula, thalamus, and prefrontal areas involved in emotion regulation. 10 , 21 , 23 , 25 , 30 , 36 Notably, 1 study found that children who would go on to develop multisite pain already exhibited differences in regional activity and structure, suggesting early indicators of widespread pain. 21 Thalamocortical dysrhythmia has been implicated in widespread pain populations and may help explain broader cortical disruptions observed in association with alpha oscillations. 22 , 28 , 41 One study demonstrated that therapeutic thalamic lesions could alleviate pain and increased PAF, directly linking thalamic dysfunction to altered cortical alpha oscillations. 36 Shedding light on this relationship between thalamic activity and alpha oscillatory activity in chronic pain populations could expand our understanding of global cortical dysfunction in this group but also better inform future clinical interventions. As a secondary analysis of an existing dataset, these findings should be interpreted within the constraints of the original study design and sample characteristics. Although groups were sex-balanced to minimize confounding, the modest and unequal sample sizes limited statistical power and generalizability, such that the study was sufficiently powered only to detect relatively large effects consistent with the robust channel-based PAF findings and the smaller, nonsignificant component-based effects. Electroencephalography data were collected in the supine position to reduce muscle activity and head motion. Although this differs from seated recordings in prior PAF–pain studies, existing evidence suggests that posture influences alpha power more than frequency, indicating minimal impact on the PAF estimates. 3 In addition, sensorimotor components were identified using an algorithm rather than visual inspection, which has been shown to offer marginal advantages in healthy populations, 7 , 14 although this comparison has not yet been made in chronic pain populations. The absence of a healthy control group is a further limitation, as it restricts direct comparison of cortical activity between pain groups and pain-free individuals. Although previous work has included healthy controls, diagnostic differences limit direct comparison with the present UCPPS cohort. 31 Future studies should, therefore, examine spatial PAF variations across additional chronic pain diagnoses and healthy populations to assess generalizability. Incorporating multimodal approaches—including structural imaging, thalamocortical connectivity, and QST—may further clarify the neural mechanisms linking PAF to widespread pain. Finally, future GICA analyses would benefit from larger, balanced samples to improve stability and interpretability. In summary, our results demonstrate that the global average PAF, but not sensorimotor component, significantly differs between widespread and localized chronic pain groups. Channel-based PAF (global average and various ROIs) better explains differences between pain groups compared with component-based PAF (sensorimotor component and GICA components). This effect reflects global cortical dysfunction in widespread compared with localized pain as channel-based methods capture a more distributed, network-level slowing of PAF consistent with global cortical dysfunction in this population while component-based methods isolate spatially independent sources, which would reflect localized dysfunction if distinctions in patient groups were identified with these methods. Furthermore, we demonstrate that pain intensity and pain duration have no effect on differences in PAF between localized and widespread pain. These results need to be further explored in other chronic pain diagnoses as well as healthy populations to create a more robust understanding of the association between PAF, pain, and sensory processing.

Supplemental digital content associated with this article can be found online at http://links.lww.com/PR9/A399 .

The authors have no conflict of interest to declare.