Background: Chronic widespread musculoskeletal pain (CWP) is a key feature of fibromyalgia and a common, disabling presentation in rheumatology. Despite this, objective biomarkers to support stratification and mechanistic understanding are lacking. Current evidence implicates altered central pain processing [1], motivating interest in neurophysiological measures. Electroencephalography (EEG) is an accessible, low-burden measure of brain activity with potential relevance to central pain mechanisms. However, EEG signals show substantial inter-individual variability influenced by age, sex, genetics [2], and shared environment, limiting interpretability in conventional case-control studies. A co-twin control design including pairs discordant for CWP provides a uniquely controlled framework to evaluate neurophysiological signatures while minimising these confounders. EEG is a high-dimensional, dynamic signal in which pain-related effects are likely to be subtle, distributed across time and channels, and not well captured by a small set of pre-defined summary metrics. Machine learning time-series methods can integrate these multivariate patterns to identify reproducible signatures that may support biomarker development for CWP.

Objectives: This co-twin control study aimed to evaluate whether resting-state EEG contains discriminative signatures of CWP in discordant twin pairs (one twin has CWP, the other doesn’t) using time-series classification, and to establish a reproducible EEG workflow feasible for biomarker discovery in TwinsUK.

Methods: Twin pairs discordant for CWP (based on criteria outlined by White et al ³ ) were recruited through the TwinsUK registry (King’s College London) in collaboration with the University of East Anglia. Eyes-closed resting-state EEG (7 minutes) was recorded using a 32-channel, salt-water, rapid-setup system. Data were minimally pre-processed to conserve brain-related signal information (0.5-60Hz band-pass; 49-51Hz notch to remove line noise) and classification used MiniROCKET (a fast time-series algorithm based on random convolutional features) from the AEON open-access toolkit. Analysis used 200-second segments (100,000 samples) and was cross-validated using 10 randomly created, class-balanced train-test splits, keeping co-twins together (pair-aware splitting). Performance was summarised by accuracy, confusion-matrix counts, sensitivity and specificity. Analyses were repeated within canonical EEG frequency bands (delta, theta, alpha, beta, gamma).

Results: The cohort comprised 36 discordant pairs (17 monozygotic, 19 dizygotic pairs), mean age 68.9 years (SD 8.0), 93% female (n=67). MiniROCKET achieved a mean accuracy of 0.62 across 10 class-balanced resamples (best resample accuracy 0.69). Overall classification outcomes were True Positives=108, True Negatives=116, False Positives=64, False Negatives=72, with mean sensitivity 0.61 and specificity 0.64. Table 1 shows band-specific accuracies, with the highest values in the alpha (0.62) and beta (0.60) bands. Together, these findings indicate a modest but above-chance discrimination of CWP status from resting-state EEG, with potentially informative signal in the alpha and beta frequency bands.

Conclusions: In a uniquely controlled discordant twin design, resting-state EEG demonstrated modest but reproducible discrimination of CWP using time-series classification with pair-aware cross-validation. These findings support the feasibility of identifying neurophysiological signatures of CWP while minimising genetic and shared environmental confounding, using a fast and low-burden EEG setup.

REFERENCES: [1] Fitzcharles MA, Cohen SP, Clauw DJ, Littlejohn G, Usui C, Hauser W. Nociplastic pain: towards an understanding of prevalent pain conditions. Lancet. 2021;397(10289):2098–2110.

[2] Smit DJA, Posthuma D, Boomsma DI, de Geus EJC. Heritability of background EEG across the power spectrum. Psychophysiology. 2005;42(6):691–697.

[3] White KP, Harth M, Speechley M, Ostbye T. Testing an instrument to screen for fibromyalgia syndrome in general population studies: the London Fibromyalgia Epidemiology Study Screening Questionnaire. J Rheumatol. 1999 Apr;26(4):880-4.

Acknowledgments: NIL.

Disclosure of Interests: None declared.