Everyone Improved, Not Equally: A Year of New-Onset Rheumatoid Arthritis
- In 158 adults with newly diagnosed rheumatoid arthritis (mean age 58.4 years, 35% male), bodily pain improved markedly over 12 months, but a higher PainDETECT score at diagnosis predicted a worse endpoint within that improvement: adjusted beta = -0.43 per PainDETECT point on the SF-36 Bodily Pain subscale, 95% CI -0.84 to -0.02, p = 0.043.
- The association held after adjustment for socio-demographics, lifestyle, comorbidities, depression and anxiety, and it was independent of inflammation: PainDETECT showed no association with C-reactive protein over the same year.
- Neuropathic-like features were present at the start of the disease, not only late in it: 91 of the 158 participants (57.6%) scored at or above the PainDETECT screening cutoff of 13, and 24.7% reached the "likely neuropathic" threshold of 19.
- In an embedded imaging substudy of 27 participants scanned at 3T, higher PainDETECT scores correlated with greater activation in the left middle and posterior insula, dorsal anterior cingulate cortex and left amygdala during a calibrated 50 N pressure stimulus to the affected wrist, contrasted against a nailbed control site, at a whole-brain threshold of Z greater than 3.1, cluster corrected at p less than 0.01.
Rheumatoid arthritis is one of the few conditions in which a peripheral pain driver can be measured, treated and largely switched off, which makes it a natural test of what pain is left over. Treat-to-target regimens bring C-reactive protein down; a sizeable minority of patients go on hurting and are then quietly relabelled as anxious, deconditioned or catastrophising. A prospective Oxford cohort, published in Pain, asked a narrower and more answerable question: does a short questionnaire completed at diagnosis predict how much pain a patient reports a year later, and does it do so independently of the inflammation everyone is busy treating?
Everyone improved, but not to the same place
Kelleher and colleagues recruited 158 adults with newly diagnosed rheumatoid arthritis in Oxfordshire (mean age 58.4 years, 35% male). Each completed the PainDETECT questionnaire at baseline. Bodily pain was measured with the SF-36 Bodily Pain subscale at baseline and at 3, 6 and 12 months, and C-reactive protein was tracked longitudinally from linked healthcare records. Linear mixed-effects models related the baseline PainDETECT score, treated as a continuous measure, to both outcomes over time.
Bodily pain improved substantially over the year. In the fully adjusted model the SF-36 bodily pain subscale stood 16.77 points above baseline by three months (on this subscale a higher score means less pain), and the gain was held at six and twelve months. Against that improving background, higher baseline PainDETECT predicted worse bodily pain throughout follow-up (adjusted beta = -0.43, 95% CI -0.84 to -0.02, p = 0.043): patients who scored higher at diagnosis ended the year worse off than their peers, not unimproved. The estimate survived adjustment for socio-demographics, lifestyle, comorbidities, depression and anxiety. Scale the coefficient and it stays modest: a 10-point difference in PainDETECT corresponds to roughly 4.3 points on a 100-point subscale, with a confidence interval running from about 0.2 to 8.4 (that arithmetic is ours, not the authors'). The result that carries more weight is the null one. Over the same twelve months, the same questionnaire showed no association with C-reactive protein at all. Two things were being measured, and they were not the same thing. Neuropathic-like features were also common at onset rather than a late consequence of years of disease: 91 of 158 participants scored at or above the screening cutoff of 13, and roughly a quarter (24.7%) reached 19.
Twenty-seven participants entered an imaging substudy. On a 3T scanner, across two ten-minute runs, an MR-compatible algometer delivered pseudorandomised blocks of 50 N pressure and light touch to the left wrist, typically affected in rheumatoid arthritis, and to the left index-finger nailbed as a control site. Contrasting pain against touch, and wrist against nailbed, at a whole-brain threshold of Z greater than 3.1 with cluster correction at p < 0.01, higher PainDETECT scores tracked stronger responses in the left middle and posterior insula, dorsal anterior cingulate cortex and left amygdala: the salience and affective end of the pain network rather than primary sensory cortex.
Where this lands in our work
The clinically useful move here is not the brain map. It is that differences in how much pain patients would still be carrying a year later became predictable, prospectively, from something the patient could report on day one. That reframes a conversation many of us inherit second-hand. When a rheumatologist says the joints look quiet and the patient says the pain has not settled as much as they expected, the standard fork is either escalate the biologic or refer to psychology with the implication that the remaining pain is manufactured. This cohort supports a third reading: a proportion of these patients arrived with a baseline symptom profile that predicted the course their pain would take, and it was not visible in their bloods.
Two cautions belong in the same breath. The imaging is cross-sectional and correlational in 27 people; stronger insula and cingulate responses to pressure are a correlate of the score, not a demonstration that the brain is generating the pain. And the inference runs in only one direction. Nothing here licenses the older and lazier reading, that pain without inflammation is not real pain; the stimulus was a calibrated 50 N, the response was measured, and the prognostic signal survived adjustment for depression and anxiety rather than being explained by them. Practically: ask about pain quality early, before two years of failed escalation, and use a positive screen to justify addressing sleep, pacing, activity and pain-specific psychological work alongside disease-modifying treatment, not as a consolation prize after it.
A pain-quality score taken on the day of diagnosis carried prognostic information that a year of inflammatory markers did not.
The cohort was recruited in a single region, Oxfordshire, so the case mix may not transfer. The imaging substudy is small, 27 of the 158, and cross-sectional. The prognostic signal is modest in standardised terms, standardised beta = -0.12, and the unstandardised confidence interval nearly touches zero. A prospective cohort of this design supports prediction rather than causation: the score identifies who is likely to do worse, not why. And PainDETECT was designed to screen for neuropathic pain rather than to measure central sensitisation, so its use as a proxy is a hypothesis the authors advance rather than a validated equivalence.