Skip to content

Biomarkers in fibromyalgia

TL;DR — After three decades of searching there is no validated diagnostic biomarker for fibromyalgia (FM), and the field's failure is structural rather than technical. Every candidate class — quantitative sensory testing (QST), fMRI classifiers, cytokine panels, metabolomic and proteomic signatures, vibrational spectroscopy, gut-derived metabolites, anti-satellite-glia IgG — has produced impressive discovery-set discrimination (typically 80–95% accuracy) and essentially no independent external validation. Two problems make this more than an ordinary translational lag. First, the reference standard is itself a symptom questionnaire (ACR criteria), so a "diagnostic" biomarker can only ever be validated against a construct it was meant to replace — the circularity problem. Second, nearly all studies compare FM patients against healthy controls rather than against the clinically relevant comparator (other chronic-pain and rheumatic patients), producing spectrum bias that inflates apparent specificity. The most-cited blood-biomarker meta-analysis found reproducible group-level differences in IL-6, IL-8, TNF-α, IFN-γ, CRP and BDNF but concluded none is specific to FM (Kumbhare 2022, PMID 34966131). The realistic near-term goal is not a diagnostic test but a stratification marker that predicts treatment response.

1. What would count as a usable FM biomarker

A biomarker earns clinical use only if it does something the existing criteria cannot. Four candidate roles, in descending order of tractability:

Role What it must do Comparator that matters Current best candidate
Stratification / predictive Sort patients into groups with different treatment response FM patients on drug A vs drug B QST profiles, none validated prospectively
Mechanistic / pharmacodynamic Move with the intervention and mediate its clinical effect Within-patient, pre/post Conditioned pain modulation (CPM), brain signature response
Confirmatory diagnostic Distinguish FM from other chronic-pain and rheumatic conditions RA, OA, SLE, chronic low back pain, ME/CFS Vibrational spectroscopy (small, single-centre)
Screening / case-finding Positive in FM before the criteria are applied, in unselected primary care Full spectrum of musculoskeletal complaints None

The single most consequential design requirement is the comparator. A test that separates FM from healthy volunteers answers a question no clinician asks. The overwhelming majority of published FM biomarker studies use healthy controls; the small minority that use disease controls (Hackshaw 2019, PMID 30523152; Yao 2023, PMID 36979691; Venerito 2026, PMID 42442924) are correspondingly more informative and correspondingly rarer.

2. The circularity problem and its consequences

FM is defined by patient-reported symptoms — widespread pain index plus symptom severity scale (see diagnostic criteria). There is no autopsy standard, no tissue lesion, no independent adjudication. Any biomarker's sensitivity and specificity are therefore computed against a questionnaire.

Three consequences follow, and they run through every section below:

  1. A perfect biomarker is unrecognisable. A marker that identified a true biological subgroup within, or across, the criteria-defined population would score as inaccurate against the criteria. The field's optimisation target penalises exactly the discovery it wants.
  2. Criteria-set drift changes the answer. Prevalence and case mix shift substantially between ACR 1990, 2010/2011 and 2016 definitions, so a biomarker validated against one criteria set is not automatically validated against another.
  3. Symptom-derived predictors will always outperform biology. In a metabolomics study using machine learning, an algorithm trained on Fibromyalgia Impact Questionnaire items classified FM with 88% balanced accuracy — and fatigue alone reached 86% — while the best 13-metabolite biological model reached 79% (Zetterman 2024, PMID 38411371). When the label is symptoms, symptoms win.

Related consequence for trials: because there is no biological entry criterion, trial populations are heterogeneous by construction, which is one driver of the placebo-response problem described in clinical trials landscape.

3. Quantitative sensory testing

QST is the most mechanistically motivated candidate class: it measures the phenomenon (augmented central pain processing) rather than a correlate of it. It is also the class with the clearest evidence that group-level differences do not become individual-level tests.

A systematic review of non-invasive markers of "human assumed central sensitization" in FM covering 78 studies and 5,234 participants found at-least-moderate evidence for a defined set of peripheral-manifestation markers — pain after-sensation decline rates, mechanical and pressure pain thresholds, sound pressure pain threshold, cutaneous silent period, slowly repeated evoked pain (SREP) sensitization and nociceptive flexion reflex threshold — and for two central-manifestation markers: CPM efficacy with pressure conditioning, and brain perfusion analysis (Smeets 2024, PMID 38073369). Notably, that review reports evidence that these markers differ between groups; it does not report validated diagnostic thresholds.

Where diagnostic accuracy has been computed, the numbers are modest. SREP — a protocol of nine suprathreshold 5-second pressure stimuli at 30-second intervals, indexed by the increase in pain ratings across stimuli — separated episodic migraine patients from healthy controls with up to 75% diagnostic accuracy, with an index cut-off of 0.5 giving 0.88 sensitivity; in the same work, conventional pain threshold, pain tolerance and temporal summation showed no significant discriminative ability (de la Coba 2021, PMID 33633294). The protocol was originally developed in FM, where it differentiated patients from both healthy individuals and rheumatoid arthritis patients — the rare disease-control design.

Practical constraints on QST as a diagnostic:

  • Test–retest and operator dependence. Thresholds are influenced by instruction, attention, expectation and examiner; effect sizes shrink with standardisation.
  • Overlap. Pressure pain thresholds are shifted downward in FM as a distribution, but the FM and control distributions overlap heavily; no single cut-point yields both usable sensitivity and usable specificity.
  • Nonspecificity. Central-sensitization-type QST profiles occur across chronic overlapping pain conditions (see comorbidities and overlap), so an abnormal QST profile localises a mechanism, not a diagnosis.
  • Small-fibre confound. Roughly half of people with FM show small-fibre pathology on skin biopsy or corneal confocal microscopy, and QST abnormalities in FM are found primarily in mechanical/pressure pain thresholds (Marshall 2025, PMID 39806197) — so an abnormal QST result may index peripheral input rather than central amplification (see peripheral pathophysiology).

QST's realistic role is stratification, not diagnosis. That role remains unproven: in the FINAL trial of low-dose naltrexone, the only QST outcome showing a between-group difference was CPM change, and sensitivity analyses showed no association between CPM change and clinical pain improvement, leading the authors to call the finding random (Bruun 2025, PMID 40214857).

4. Neuroimaging classifiers

The landmark paper remains López-Solà 2017 (PMID 27583567): 37 FM patients and 35 matched healthy controls, fMRI during painful pressure and during non-painful multisensory (visual–auditory–tactile) stimulation. Three brain patterns were combined — the pre-existing Neurologic Pain Signature, a new "FM-pain" classifier, and a "multisensory" classifier trained on non-painful stimulation — to classify patients versus controls with 92% sensitivity and 94% specificity in out-of-sample individuals. Mechanistically the result is informative: augmented responses in insula/operculum and self-referential medial prefrontal regions, reduced responses in primary/secondary sensory cortices, basal ganglia and cerebellum — a signature of altered sensory integration rather than of amplified nociception alone (see central pathophysiology).

Subsequent work has reproduced the approach rather than the classifier. Combined resting-state functional connectivity and structural MRI features in 26 FM patients and 30 healthy controls, with recursive feature elimination, reached accuracy 0.95 and AUC 0.95 for the combined model (rs-FC alone: accuracy 0.91, AUC 0.93; structural alone: 0.86/0.88) (Thanh Nhu 2022, PMID 36551758). A review of brain-imaging classifiers across chronic musculoskeletal pain conditions places the achievable range at 70–92% accuracy and flags the scientific, practical and ethical problems of clinical deployment (Boissoneault 2017, PMID 28144827).

What is missing is uniform across these papers and is the reason none has entered practice:

  • No independent external validation. In this session's PubMed searching we found no study applying the López-Solà signature, unchanged and pre-specified, to an independent FM cohort collected by a different group on different scanners. Reported accuracies are cross-validated within sample, which is not the same thing.
  • Healthy-control comparator. Discrimination is against pain-free controls, not against chronic low back pain, RA or ME/CFS.
  • Sample sizes of 20–40 per arm with feature spaces of thousands of voxels — the regime in which cross-validated accuracy is known to be optimistically biased.
  • Cost and access. Even a validated fMRI classifier would cost orders of magnitude more than the questionnaire it replaces.

5. Blood-based candidates

5.1 Cytokines and the meta-analytic picture

The definitive synthesis is a systematic review and meta-analysis of 54 studies (40 meta-analysed) comparing circulating immune mediators in FM patients and healthy controls. Patients had significantly lower IL-1β and higher IL-6, IL-8, TNF-α, IFN-γ, CRP and BDNF. The authors' conclusion is the key sentence in this literature: the evidence does not support these as specific biomarkers of FM, though individual markers may help identify coexisting pathology (Kumbhare 2022, PMID 34966131). Effect sizes are group-level; distributions overlap; the direction of some markers is inconsistent across studies.

Stress-axis markers fare no better. A meta-analysis of 47 studies (1,465 FM, 1,192 controls) found no main effect of FM on blood cortisol, ACTH, CRH or epinephrine; salivary and urinary cortisol were lower and blood norepinephrine higher in FM, but with high heterogeneity and significant evidence of publication bias, leading the authors to decline to assert abnormal HPA-axis function in FM (Beiner 2023, PMID 36728497).

A narrower systematic review of shared inflammatory pathways across FM, depression and autoimmune disease reaches the same place from a different direction: cytokine profiles and platelet serotonin activity are called "emerging" biomarkers requiring validation in large multicentre studies (Sedda 2025, PMID 40002916).

5.2 The commercial cytokine-panel test (FM/a) and what is actually published

A commercially marketed blood test for FM (the FM/a test; the marketer identification — EpicGenetics, US — comes from the test's public marketing and was not confirmed in any source retrieved this session [unverified]) rests on a published cytokine-response assay. The underlying study compared plasma and peripheral blood mononuclear cells from 110 patients with a clinical FM diagnosis and 91 healthy donors; PBMC were cultured overnight in medium alone or with mitogenic activators (PHA, or PMA plus ionomycin), and IFN-γ, IL-5, IL-6, IL-8, IL-10, MIP-1β, MCP-1 and MIP-1α were measured by bead-array multiplex. Stimulated cytokine concentrations were lower in FM samples than in controls, with decrements ranging from 1.5-fold (MIP-1β) to 10.2-fold (IL-6) under PHA challenge and 1.8–4-fold under PMA challenge. The authors concluded that cell-mediated immunity is impaired in FM and that the assay "can offer a diagnostic methodology" when combined with clinical patterns (Behm 2012, PMID 23245186). The paper's competing-interests statement, checked in the full text retrieved this session, declares no competing interests — so whatever commercial relationship exists between the study team and the test's marketer is not documented in the publication itself.

What that publication does and does not establish:

  • It reports group differences, not diagnostic accuracy: no ROC curve, no sensitivity, no specificity, no positive or negative predictive value is presented in the abstract, and no cut-point is defined.
  • The comparator is healthy donors — the spectrum-bias problem in its purest form. Reduced mitogen-stimulated cytokine production is not FM-specific; it is seen with corticosteroids, intercurrent illness, smoking, and sample-handling delay.
  • There is no independent replication by an unaffiliated group that we could verify by PubMed search in this session, and no prospective study in the intended-use population (patients being evaluated for FM in primary or specialty care).
  • Pre-test probability determines the answer. Even a test with 90% sensitivity and 90% specificity applied in a population with 5% FM prevalence yields a positive predictive value near 32% — most positives would be false. The commercial framing of the test as diagnostic confirmation is not supported by the published accuracy data, because published accuracy data do not exist.

Published expert critiques of the specific commercial test could not be verified by PubMed search in this session and are marked [unverified]; the evidentiary gaps above are stated from the primary publication itself rather than from a secondary critique.

5.3 Proteomics

A systematic review of 10 observational proteomic studies identified 3,328 proteins of which 145 were differentially expressed between FM patients and controls, across plasma, serum, cerebrospinal fluid and saliva, with control groups including both healthy individuals and patients with inflammatory and non-inflammatory pain. Recurrent hits were transferrin, α-, β- and γ-fibrinogen chains, profilin-1, transaldolase, PGAM1, apolipoprotein-C3, complement C4A and C1QC, immunoglobulin fragments and acute-phase reactants — a signature the authors read as complement/coagulation dysregulation, iron metabolism and oxidative stress. Correlations with pain sensitivity or quality-of-life scores were weak apart from transferrin and α2-macroglobulin with moderate-to-severe pain (Gkouvi 2024, PMID 38652420). The heterogeneity of matrices, platforms and control groups across ten studies is itself the finding: there is no consensus FM proteome.

5.4 Metabolomics

The most methodologically explicit study enrolled 54 FM patients and 31 healthy controls and sampled metabolites at baseline and after two standardised stressors (oral glucose tolerance test and cardiopulmonary exercise test). Supervised learning reduced 77 metabolomic markers to 13 key markers, which identified FM in held-out cases with 79% accuracy; 5-hydroxyindole-3-acetic acid and glutamine correlated with fatigue severity; patients differed from controls in tyrosine and purine pathways at baseline and in the pyrimidine pathway after challenge (Zetterman 2024, PMID 38411371). The same paper contains the field's most useful negative control, described in §2: symptom-only models beat the biology.

A multi-omic study from a commercial laboratory analysed plasma proteome and faecal metagenome in 199 FM patients and 43 environmentally paired controls, reporting 30 differentially abundant proteins and 19 differentially abundant taxa whose integration into an algorithm discriminated cases from controls, with GAPDH highlighted (Durán-González 2025, PMID 41113645). The registered parent study (NCT05921409, "FIBROKIT", Pronacera Therapeutics, planned n = 250: 206 patients and 44 healthy volunteers, women aged 40–59, with an olive-oil-supplemented Mediterranean diet arm) makes the comparator explicit: healthy volunteers, with exclusion of inflammatory, autoimmune, gastrointestinal, cardiovascular and metabolic disease. That exclusion list removes precisely the conditions from which a diagnostic test would need to distinguish FM. The 199:43 case:control ratio, single-centre design and commercial sponsorship all require independent replication before the discrimination claim can be assessed.

5.5 microRNAs

The circulating-miRNA literature in FM recapitulates the field's pattern in miniature: small discovery studies with strong within-sample numbers, and a specificity problem as soon as disease controls appear.

  • Discovery signatures in tiny samples. A PBMC microarray study in 11 FM patients and 10 controls proposed a five-miRNA downregulated signature (miR-223-3p, miR-451a, miR-338-3p, miR-143-3p, miR-145-5p), with ~20% of all analysed miRNAs downregulated ≥2-fold — read by the authors as possible global dysregulation of miRNA synthesis — and no correlation with cardinal symptoms; validation in larger groups was explicitly required (Cerdá-Olmedo 2015, PMID 25803872). A serum-and-saliva panel study in 14 FM patients built a five-miRNA linear model reaching 100% sensitivity and 83.3% specificity (Masotti 2017, PMID 27796750) — numbers of the kind that n = 14 healthy-control designs reliably produce and external validation reliably deflates.
  • The disease-control test. The largest sequencing study profiled blood and keratinocyte miRNAs and tRNA fragments in 53 FM patients, 34 healthy controls and 15 disease controls (major depression with chronic physical pain). ROC analysis of blood candidates (miR-148a-3p, miR-182-5p, one tRF) separated FM from healthy controls — but validated miR-182-5p and miR-576-5p were even higher in the disease controls than in FM; only a tRNA fragment (tRF-20-40KK5Y93) was selectively increased in FM (Erbacher 2025, PMID 39679614). As with cytokines and anti-SGC IgG, the candidates index distress or chronic illness, not FM.
  • Differential diagnosis rather than case-finding may be the class's more realistic role: an 11-miRNA panel with machine learning discriminated ME/CFS from FM and from comorbid ME/CFS+FM (Nepotchatykh 2023, PMID 36732593).

No externally validated circulating-miRNA diagnostic exists, and no study has tested a locked miRNA signature prospectively in an intended-use population.

6. Vibrational spectroscopy

The most striking published discrimination in the FM biomarker literature comes from infrared and Raman spectroscopy of dried bloodspots — and it is also the clearest illustration of why single-centre discrimination is not validation.

In the index study, bloodspots from 50 FM, 29 rheumatoid arthritis, 19 osteoarthritis and 23 SLE patients were analysed by portable FT-IR and FT-Raman microspectroscopy with pattern-recognition analysis. Spectral signatures clustered participants into disease classes with no misclassification (p < 0.05, interclass distances > 2.5), and spectra correlated with FIQR severity at r = 0.95 (IR) and r = 0.83 (Raman). Protein backbones and pyridine-carboxylic acids dominated the discrimination (Hackshaw 2019, PMID 30523152). A follow-up using a portable FT-IR instrument and four sample-preparation methods in 122 FM patients and 70 patients with related rheumatologic disorders (SLE 17, RA 43, OA 10) reported OPLS-DA classification with Rcv > 0.93 and "excellent sensitivity and specificity", with peptide backbones and aromatic amino acids driving separation (Yao 2023, PMID 36979691).

Strengths worth stating plainly: this line uses disease controls, not healthy volunteers; the assay is cheap, fast and deployable at point of care; and it produced a severity correlation, not only a binary split.

Weaknesses equally plainly:

  • Perfect or near-perfect separation in a sample of this size with a high-dimensional spectral feature space is the classic signature of overfitting, and both papers report internal cross-validation rather than a locked model tested on an independent cohort.
  • Batch and site effects. Bloodspot cards, storage time, humidity and instrument drift all shift spectra; if FM and comparator samples were collected in different clinics or periods, the classifier may be learning provenance.
  • The identified discriminants (protein backbones, aromatic amino acids) are not disease-specific chemistry; they are generic macromolecular features.
  • No prospective, blinded, multi-site validation has been published that we could verify in this session.

7. Autoantibody and immune candidates

The autoantibody line is covered in depth in autoimmunity and inflammation; only its biomarker implications belong here.

Passive transfer of IgG from FM patients into mice produced mechanical and cold hypersensitivity, reduced grip strength, reduced locomotor activity and loss of intraepidermal innervation, while IgG-depleted patient serum and healthy-control IgG did not; patient IgG labelled satellite glial cells and neurons in mouse dorsal root ganglia and also bound human DRG (Goebel 2021, PMID 34196305). This is the strongest existing evidence for a causal circulating factor in FM, and it immediately suggests anti-satellite-glial-cell (anti-SGC) IgG as a candidate biomarker.

The specificity problem arrived quickly. In a study comparing pooled IgG from patients with post-acute COVID-19 syndrome (PACS) with high pain and fatigue, from people recently recovered from acute COVID-19, and from FM patients: FM-IgG reproduced the mouse phenotype, PACS-IgG did not, yet both PACS-IgG and recently-recovered-COVID IgG stained satellite-glial-cell-enriched cultures strongly positive (Berwick 2025, PMID 40408228). Anti-SGC reactivity is therefore not specific to FM and cannot serve as a diagnostic marker on its own; the functional (pronociceptive) property, not the binding, is what distinguishes FM IgG — and a passive-transfer bioassay in mice is not a clinical test.

The mechanism has since been extended: FM IgG sensitises Aβ low-threshold mechanoreceptors to mechanical and cold stimuli, matching the tingling and cold-evoked pain that patients report (Israel 2025, PMID 40898777).

Separately, autoantibodies against trisulfated heparin disaccharide (TS-HDS) and fibroblast growth factor receptor 3 (FGFR-3) are found in a substantial fraction of cryptogenic small-fibre neuropathy cases that are "often misdiagnosed as other conditions including fibromyalgia" (Zeidman 2021, PMID 34019003). These are best understood not as FM biomarkers but as markers that identify patients who should not be labelled FM — a diagnostically useful role, and a reminder that the FM label absorbs identifiable disease.

8. Symptom-derived and record-derived classifiers

A distinct and often-overlooked category: models built from data that already exist. Random-forest modelling of electronic medical records identified ten variables (counts of laboratory/diagnostic test orders, outpatient and office visits, opioid prescriptions, total medications, non-opioid pain medications, emergency-room visits, musculoskeletal conditions, and age) that predicted a recorded FM diagnosis with AUC 0.810 in an independent test set (Emir 2015, PMID 26089700). A questionnaire-based triage classifier reduced from 67 to 28 items reached AUC 0.81 for identifying FM among newly referred rheumatology patients, on external validation (Maarseveen 2025, PMID 41248315). A multimodal AI system combining video descriptors and psycholinguistic transcript features distinguished FM from other rheumatic and musculoskeletal diseases in 50 FM and 50 non-FM patients with 91% accuracy, 94% sensitivity, 88% specificity and AUC 0.96 — with classification thresholds optimised post hoc by Youden's J index, which the authors themselves state requires external validation before deployment (Venerito 2026, PMID 42442924).

These matter for two reasons. First, they set the bar any biological marker must clear: an AUC of 0.81 is available from a questionnaire and free from an EHR. Second, they demonstrate that "diagnostic accuracy for FM" is achievable without touching biology at all — which is precisely what one expects when the reference standard is symptomatic.

9. Candidate summary table

Candidate class Best reported discrimination Comparator used External validation Verdict
fMRI multi-pattern signature 92% sens / 94% spec (López-Solà 2017, PMID 27583567) Healthy controls None found Mechanistically important, not a test
rs-FC + structural MRI ML Accuracy 0.95, AUC 0.95, n=56 (PMID 36551758) Healthy controls None found Overfitting-prone sample size
QST / SREP ~75% accuracy (PMID 33633294) Healthy controls; RA in FM work Partial Mechanism marker, not diagnostic
Cytokine panels (research) Group differences only (PMID 34966131) Healthy controls Meta-analysed; not specific Not a diagnostic marker
Commercial cytokine assay (FM/a) Fold-change differences only (PMID 23245186) Healthy donors None found Accuracy data not published
HPA-axis markers No main effect (PMID 36728497) Healthy controls Meta-analysed; publication bias Negative
Proteomics 145/3,328 proteins differential (PMID 38652420) Mixed Inconsistent across 10 studies No consensus signature
Metabolomics 79% accuracy, 13 markers (PMID 38411371) Healthy controls None found Beaten by fatigue item alone
Vibrational spectroscopy Near-perfect clustering (PMIDs: 30523152, 36979691) RA / OA / SLE None found Best comparator, weakest validation
Circulating miRNAs 100% sens / 83% spec, n=14 (PMID 27796750) Healthy controls; depression in one study Candidates higher in disease controls (PMID 39679614) Not FM-specific
Microbiome ML AUC 87.8% (PMID 31219947) Healthy controls Contradicted (PMID 42550534) See omics
Anti-SGC IgG Binding not FM-specific (PMID 40408228) PACS, post-COVID Refuted as specific marker Mechanism, not marker
EHR / questionnaire ML AUC 0.81 (PMIDs: 26089700, 41248315) Clinical population Yes (PMID 41248315) Sets the bar for biology

10. Why nothing has validated: a diagnosis of the field

  1. Spectrum bias. Healthy-control designs inflate specificity because healthy people differ from FM patients in deconditioning, sleep, medication, BMI, smoking and distress — all of which move cytokines, metabolites and brain signals.
  2. No gold standard, hence circularity (§2).
  3. Small samples with high-dimensional features. Cross-validated accuracy in n = 25–60 per arm with thousands of features is systematically optimistic; the field reports it as though it were external validation.
  4. Publication bias. Demonstrated directly for HPA-axis markers in FM (Beiner 2023, PMID 36728497) and plausible across every other class.
  5. Population heterogeneity. FM as defined almost certainly contains several biologically distinct groups (a peripheral small-fibre subgroup, an IgG-mediated subgroup, a predominantly central subgroup); a single marker averaged across them is diluted by construction.
  6. No intended-use studies. Almost nothing has been tested prospectively in the population where it would be used — patients presenting with widespread pain of unknown cause, before diagnosis.

Open questions

  • Does any published FM classifier survive locked-model, blinded, multi-site external validation? None was identified in this session's searches for the fMRI signature (PMID 27583567), the spectroscopy signature (PMIDs: 30523152, 36979691) or the metabolomic signature (PMID 38411371) — the question is open because the study has not been done, not because it failed.
  • Can any candidate discriminate FM from ME/CFS, chronic low back pain and inflammatory arthritis with concomitant nociplastic pain, rather than from healthy controls? Only the spectroscopy line and one AI system (PMID 42442924) have used disease comparators at all.
  • Is anti-SGC IgG functional reactivity (pronociceptive in passive transfer) measurable by an in vitro assay that could become a clinical test, given that binding alone is non-specific (PMID 40408228)?
  • Does QST-defined phenotype predict differential treatment response? The FINAL trial's CPM result was interpreted by its own authors as a chance finding uncorrelated with clinical improvement (PMID 40214857), so the stratification hypothesis remains untested rather than refuted.
  • What is the diagnostic accuracy of the marketed FM/a test in its intended-use population? The foundational publication (PMID 23245186) reports fold-change differences against healthy donors and no accuracy metrics; no independent evaluation was locatable.
  • Do circulating miRNAs discriminate FM from disease controls? In the one study with a disease-control arm, the leading blood miRNA candidates were even higher in depression-with-chronic-pain controls than in FM, and only a tRNA fragment was FM-selective (PMID 39679614) — is tRF-20-40KK5Y93 replicable, and against which comparators?
  • Given that fatigue alone classified FM at 86% accuracy while the best biological model reached 79% (PMID 38411371), what incremental value must a biological marker demonstrate over free symptom data before it is worth developing?

References

  1. Behm FG, Gavin IM, Karpenko O, et al. Unique immunologic patterns in fibromyalgia. BMC Clin Pathol. 2012;12:25. PMID 23245186
  2. Beiner E, Lucas V, Reichert J, et al. Stress biomarkers in individuals with fibromyalgia syndrome: a systematic review with meta-analysis. Pain. 2023;164:1416-1427. PMID 36728497
  3. Berwick RJ, Sahbaie P, Kenny G, et al. Postacute COVID-19 syndrome and fibromyalgia syndrome are associated with anti-satellite glial cell IgG serum autoantibodies but only fibromyalgia syndrome serum-IgG is pronociceptive. Pain. 2025;166:e397-e408. PMID 40408228
  4. Boissoneault J, Sevel L, Letzen J, Robinson M, Staud R. Biomarkers for musculoskeletal pain conditions: use of brain imaging and machine learning. Curr Rheumatol Rep. 2017;19:5. PMID 28144827
  5. Bruun KD, Christensen R, Amris K, et al. Effect of naltrexone on spinal and supraspinal pain mechanisms and functional capacity in women with fibromyalgia: exploratory outcomes from the randomized placebo-controlled FINAL trial. CNS Drugs. 2025;39:685-692. PMID 40214857
  6. Cerdá-Olmedo G, Mena-Durán AV, Monsalve V, Oltra E. Identification of a microRNA signature for the diagnosis of fibromyalgia. PLoS One. 2015;10:e0121903. PMID 25803872
  7. de la Coba P, Bruehl S, Reyes del Paso GA. Slowly repeated evoked pain (SREP) as a central sensitization marker in episodic migraine patients. Sci Rep. 2021;11:4582. PMID 33633294
  8. Durán-González E, Ramírez-Tejero JA, Pérez-Sánchez M, et al. Fibromyalgia diagnosis from a multi-omics approach: a gut feeling. Front Microbiol. 2025;16:1641185. PMID 41113645
  9. Emir B, Masters ET, Mardekian J, Clair A, Kuhn M, Silverman SL. Identification of a potential fibromyalgia diagnosis using random forest modeling applied to electronic medical records. J Pain Res. 2015;8:277-88. PMID 26089700
  10. Erbacher C, Vaknine-Treidel S, Madrer N, et al. Altered blood and keratinocyte microRNA/transfer RNA fragment profiles related to fibromyalgia syndrome and its severity. Pain. 2025;166:1641-1652. PMID 39679614
  11. Gkouvi A, Tsiogkas SG, Bogdanos DP, Gika H, Goulis DG, Grammatikopoulou MG. Proteomics in patients with fibromyalgia syndrome: a systematic review of observational studies. Curr Pain Headache Rep. 2024;28:565-586. PMID 38652420
  12. Goebel A, Krock E, Gentry C, et al. Passive transfer of fibromyalgia symptoms from patients to mice. J Clin Invest. 2021;131:e144201. PMID 34196305
  13. Hackshaw KV, Aykas DP, Sigurdson GT, et al. Metabolic fingerprinting for diagnosis of fibromyalgia and other rheumatologic disorders. J Biol Chem. 2019;294:2555-2568. PMID 30523152
  14. Israel MR, Berwick R, Vastani N, et al. Aβ low threshold mechanoreceptors contribute to sensory abnormalities in fibromyalgia. Brain. 2025;148:4016-4029. PMID 40898777
  15. Kumbhare D, Hassan S, Diep D, et al. Potential role of blood biomarkers in patients with fibromyalgia: a systematic review with meta-analysis. Pain. 2022;163:1232-1253. PMID 34966131
  16. López-Solà M, Woo CW, Pujol J, et al. Towards a neurophysiological signature for fibromyalgia. Pain. 2017;158:34-47. PMID 27583567
  17. Maarseveen TD, Reimann F, Al Hasan A, et al. Evaluating the accuracy of the Frysian Questionnaire for Differentiation of Musculoskeletal Complaints for triage of musculoskeletal diseases: algorithm development and validation study. JMIR Med Inform. 2025;13:e77345. PMID 41248315
  18. Marshall A, Elshafei M, Preston FG, et al. Small fibre pathology in fibromyalgia: a review. Pain Ther. 2025;14:461-478. PMID 39806197
  19. Masotti A, Baldassarre A, Guzzo MP, Iannuccelli C, Barbato C, Di Franco M. Circulating microRNA profiles as liquid biopsies for the characterization and diagnosis of fibromyalgia syndrome. Mol Neurobiol. 2017;54:7129-7136. PMID 27796750
  20. Minerbi A, Gonzalez E, Brereton NJB, et al. Altered microbiome composition in individuals with fibromyalgia. Pain. 2019;160:2589-2602. PMID 31219947
  21. Nepotchatykh E, Caraus I, Elremaly W, et al. Circulating microRNA expression signatures accurately discriminate myalgic encephalomyelitis from fibromyalgia and comorbid conditions. Sci Rep. 2023;13:1896. PMID 36732593
  22. Sedda S, Cadoni MPL, Medici S, et al. Fibromyalgia, depression, and autoimmune disorders: an interconnected web of inflammation. Biomedicines. 2025;13:503. PMID 40002916
  23. Smeets Y, Soer R, Chatziantoniou E, et al. Role of non-invasive objective markers for the rehabilitative diagnosis of central sensitization in patients with fibromyalgia: a systematic review. J Back Musculoskelet Rehabil. 2024;37:525-584. PMID 38073369
  24. Thanh Nhu N, Chen DY, Kang JH. Identification of resting-state network functional connectivity and brain structural signatures in fibromyalgia using a machine learning approach. Biomedicines. 2022;10:3002. PMID 36551758
  25. Venerito V, Lopalco G, Morrone M, et al. fAI-BRO: a multimodal AI decision-support system to address diagnostic delay in fibromyalgia syndrome. RMD Open. 2026;12:e006995. PMID 42442924
  26. Yao S, Bao H, Nuguri SM, et al. Rapid biomarker-based diagnosis of fibromyalgia syndrome and related rheumatologic disorders by portable FT-IR spectroscopic techniques. Biomedicines. 2023;11:712. PMID 36979691
  27. Zeidman LA, Kubicki K. Clinical features and treatment response in immune-mediated small fiber neuropathy with trisulfated heparin disaccharide or fibroblast growth factor receptor 3 antibodies. J Clin Neuromuscul Dis. 2021;22:192-199. PMID 34019003
  28. Zetterman T, Nieminen AI, Markkula R, Kalso E, Lötsch J. Machine learning identifies fatigue as a key symptom of fibromyalgia reflected in tyrosine, purine, pyrimidine, and glutaminergic metabolism. Clin Transl Sci. 2024;17:e13740. PMID 38411371