Skip to content

López-Solà M, Woo CW, Pujol J, Deus J, Harrison BJ, Monfort J, Wager TD. Towards a neurophysiological signature for fibromyalgia. Pain. 2017;158:34-47. PMID 27583567

One-paragraph summary

Thirty-seven patients with fibromyalgia (FM) and 35 matched healthy controls underwent functional MRI during two challenges: painful pressure, and non-painful multisensory (visual–auditory–tactile) stimulation. Machine-learning techniques were used to derive a brain-based FM signature. Given identical painful stimuli, FM patients showed greater responses on the previously published Neurologic Pain Signature (NPS). A newly trained pain-related classifier ("FM-pain") revealed augmented responses in sensory-integration (insula/operculum) and self-referential (e.g. medial prefrontal) regions with reduced responses in lateral frontal cortex. A separate "multisensory" classifier trained on the non-painful sensory stimulation revealed augmented responses in insula/operculum, posterior cingulate and medial prefrontal regions, with reduced responses in primary and secondary sensory cortices, basal ganglia and cerebellum. Combining activity across the three patterns classified patients versus controls with 92% sensitivity and 94% specificity in out-of-sample individuals. Enhanced NPS responses partly mediated mechanical hypersensitivity and correlated with depression and disability; FM-pain and multisensory responses correlated with clinical pain (all P uncorrected < 0.05).

Key findings

  • 92% sensitivity / 94% specificity for FM versus healthy controls using combined NPS, FM-pain and multisensory pattern responses, evaluated out-of-sample.
  • The multisensory classifier is the conceptually important result: it was trained on non-painful visual–auditory–tactile stimulation and still discriminated FM. This locates the abnormality in generalised sensory processing rather than in nociception specifically, consistent with the clinical phenomenology of sensory hypersensitivity in FM.
  • The pattern is directional, not simply "more brain activity": augmented insula/operculum, posterior cingulate and medial prefrontal responses coexist with reduced primary/secondary sensory cortex, basal ganglia and cerebellar responses.
  • Different patterns tracked different clinical dimensions: NPS response related to mechanical hypersensitivity, depression and disability; FM-pain and multisensory responses related to clinical pain.
  • The paper explicitly frames these as candidate objective neural targets for therapeutic interventions and as a framework for assessing therapeutic mechanism and predicting treatment response at the individual level — the stratification role, not the diagnostic role.

Limitations

  • Healthy-control comparator. Discrimination is against pain-free individuals, not against other chronic-pain conditions (chronic low back pain, rheumatoid arthritis, ME/CFS) — the comparator that a diagnostic test would actually face. Spectrum bias inflates apparent specificity.
  • n = 37 vs 35 with a very high-dimensional feature space. Out-of-sample cross-validation within a single dataset is a much weaker guarantee than an independent cohort on different scanners with a locked model; accuracy in this regime is systematically optimistic.
  • Single site, single scanner. Site and acquisition effects are unmeasured and could contribute to separation.
  • Uncorrected p-values for the brain–symptom correlations (stated as P uncorrected < 0.05 in the paper), which are the claims linking the signature to clinical meaning.
  • Medication and comorbidity. FM patients differ from healthy controls in analgesic and antidepressant exposure, sleep, deconditioning and distress, all of which affect BOLD signal.
  • No independent external validation. In PubMed searching for this curation session, no study was found applying this signature unchanged and pre-specified to an independent FM cohort collected by a different group. The absence is the single most important fact about this paper's clinical status.
  • Cost. Even a validated fMRI classifier would cost orders of magnitude more than the questionnaire it would replace.

Why it matters

This is the reference point for every subsequent FM neuroimaging classification claim, and it remains the field's best-argued case that FM has a characterisable brain phenotype at the level of individual patients rather than group means.

Its intellectual contribution is the multisensory classifier. Before this paper, the dominant framing was augmented pain processing; López-Solà showed that a classifier trained on non-painful sight, sound and touch also separates FM from controls, which reframes the condition as a disorder of sensory integration and salience rather than of nociceptive gain alone. That reframing runs directly into central pathophysiology and provides the mechanistic content behind the "nociplastic pain" concept described in history and nosology.

Its methodological legacy is more ambiguous. Subsequent work has reproduced the approach — resting-state connectivity plus structural features reaching accuracy 0.95 and AUC 0.95 in 26 FM patients and 30 controls (Thanh Nhu 2022, PMID 36551758) — rather than validating the classifier. A review of the field places achievable accuracies at 70–92% across chronic musculoskeletal pain conditions while flagging the scientific, practical and ethical obstacles to clinical use (Boissoneault 2017, PMID 28144827). Nine years on, no FM brain signature is in clinical use, no external validation has been published, and the practical bar has risen: symptom- and record-based classifiers now reach AUC ≈ 0.81 essentially for free (Emir 2015, PMID 26089700; Maarseveen 2025, PMID 41248315), so an expensive imaging biomarker must justify itself against that baseline.

The honest reading is that this paper's enduring value is mechanistic and its diagnostic promise is unfulfilled — and that the reason is not that anyone tried to validate it and failed, but that the validation study has never been done.

Cited by wiki pages

  • biomarkers