Rheumatoid arthritis — open research questions¶
Last curated: 2026-11-01. Stable IDs persist across updates; answered questions should be marked resolved rather than renumbered. Tier 1 questions could change practice and are designable now. Tier 2 questions require enabling methods, longer follow-up or stronger causal foundations.
Tier 1 — practice-changing and designable now¶
OQ-1 — Who should receive pharmacologic prevention?¶
ARIAA and APIPPRA/ALTO show that abatacept can reduce or delay progression in enriched high-risk populations, while five-year TREAT EARLIER data suggest that benefit may depend sharply on serologic and absolute-risk strata (Rech 2024, PMID 38364841; Cope 2026, PMID 41576971; Mulligen 2026, PMID 42392130). External validation of 14 published risk-prediction rules in APIPPRA found poor discrimination for 10 of 14 (C-statistic <0.6) (Jasenecova 2026, PMID 42676706). Trial-design reviews now emphasize EULAR/ACR risk-stage criteria, ACPA-positive versus ACPA-negative differences, and endpoints beyond onset (van der Helm-van Mil 2026, PMID 42751359). Activated naive B cells with a senescent-like program expanded before onset in ACPA-positive progressors across independent cohorts (Wu 2026, PMID 42875583), and RF positivity in 469,036 UK Biobank participants proved genetically distinct from RA (Hocaoğlu 2026, PMID 42378266); neither has yet been tested as a calibrated selector. The 2025 Argentine recommendations advise against treating asymptomatic high-risk people without clinical arthritis while supporting methotrexate in seropositive imaging-detected subclinical arthritis (Schneeberger 2026, PMID 42767235). A multicenter trial should compare calibrated risk thresholds, overtreatment, adverse events and drug-free non-progression.
OQ-2 — How long must a preventive effect persist to count as prevention?¶
Temporary suppression and delayed onset can still be valuable, but they are not immune tolerance. ARIAA used six months of treatment; after one year of APIPPRA abatacept, ALTO found a diminishing 4.9-month arthritis-free-survival difference at four years and no persistent disease-activity or patient-reported-outcome difference (Rech 2024, PMID 38364841; Cope 2026, PMID 41576971). Follow-up should measure cumulative DMARD exposure and lifetime disease burden, not onset alone.
OQ-3 — Can synovial biopsy choose a biologic mechanism prospectively?¶
R4RA proved multicenter biopsy-stratified randomization is feasible, and RNA B-cell signatures were more informative than simple histology (Humby 2021, PMID 33485455; Rivellese 2022, PMID 35589854). A newer blood functional/proteomic classifier relates JAK- and TNF-inhibitor signatures to response, but its full text states that independent external validation is lacking and observational allocation prevents individual counterfactual inference (Corrales 2026, PMID 42907925). The next trial should randomize a locked biomarker-guided strategy versus clinical selection and test utility, cost and biopsy acceptability.
OQ-4 — Which patients can taper without flare?¶
RETRO and TARA show frequent relapse after reduction/withdrawal and no simple universal sequence advantage (Tascilar 2021, PMID 38297524; van Mulligen 2020, PMID 32482645). A 348-person 24-week trial found structured b/tsDMARD tapering non-inferior to continuation for DAS28-ESR LDA (RD −1.43%, 95% CI −8.09 to 5.23) (Park 2026, PMID 42839723), while a 5,262-person synthesis still found higher flare risk with tapering (RR 1.56) and withdrawal (RR 2.23) and a network ranked stopping bDMARDs worst (Zhang 2026, PMID 42836514; Tandi 2026, PMID 42843067). The Canadian living update conditionally permits tapering either conventional or advanced therapy first only within a gradual, monitored strategy with reliable access to care and prompt medication reinstatement after flare (Nikolic 2026, PMID 41921998). A pairwise synthesis of discontinuation versus dose reduction or maintenance found complete bDMARD stop more than doubled flare (RR 2.13, 95% CI 1.74–2.61) with 87.5% recapture (82.4–91.6) at a median 12.4 weeks, and a lower relative flare risk in early than established RA (Lee 2026, PMID 42269230). A class-specific 15-RCT review with a February 2024 search found JAK-inhibitor tapering reduced low-disease-activity maintenance (RR 0.83) while anti-TNF tapering left LDA unchanged but still increased flare (Rotea-Salvo 2026, PMID 42963167). A longer biomarker-stratified taper trial should include time to flare, recapture, structural damage, total drug exposure and access-dependent rescue delays. NCT07802678 is testing csDMARD discontinuation on TNF-inhibitor background.
OQ-5 — What is the absolute JAK-inhibitor excess risk in low-risk patients?¶
ORAL Surveillance enrolled people aged ≥50 with cardiovascular risk and found tofacitinib noninferiority failure versus TNF inhibitors for MACE and cancer (Ytterberg 2022, PMID 35081280). A propensity-matched cohort of 1,304 adults aged ≥65 did not show higher three-year MACE, VTE, malignancy or mortality on JAK versus TNF inhibitors (Ibis 2026, PMID 42823079); that observational null does not rebut the randomized enriched comparison. An 18,757-pair TriNetX target-trial emulation in ORAL Surveillance-like patients (age ≥50, ≥1 cardiovascular risk factor) also found no MACE excess (HR 1.002, 95% CI 0.923–1.088) but a secondary-outcome all-cause mortality excess (HR 1.158, 1.058–1.268; E-value 1.586) not reproduced in any agent-specific cohort (McClenahan 2026, PMID 42877103). Two observational MACE nulls against one randomized noninferiority failure sharpen the question rather than answer it: the discordance may reflect channeling, EHR outcome ascertainment, or true effect modification. CARDINAL-RA explicitly excludes cardiovascular death from its primary MACE composite and uses a post-initiation landmark, so its full-text methods establish an endpoint/design mismatch rather than a like-for-like replication (McClenahan 2026, PMID 42877103). Large active-comparator analyses should prespecify age, smoking, ASCVD, cancer and VTE strata and report risk differences per 1,000 person-years.
OQ-6 — Is the JAK safety signal molecule-, dose- or class-specific?¶
Trial and registry programs differ, while biochemical selectivity does not map simply to clinical effects (Traves 2021, PMID 33741556; Russell 2023, PMID 37247942). A seven-study real-world meta-analysis found lower discontinuation with JAK1-selective than pan-JAK agents (HR 0.72, 95% CI 0.61–0.85) from effectiveness, not from fewer adverse-event or infection stops (Lee 2026, PMID 42034889). That observational retention gradient is not a randomized molecule-versus-class safety test. Head-to-head evidence is unlikely; harmonized individual-participant analyses and negative controls are designable.
OQ-7 — Which RA patients should undergo ILD screening HRCT?¶
Age, sex, smoking, serology and genetics enrich risk, but no model has demonstrated that screening improves outcomes (Sullivan 2024, PMID 39320427; Yao 2025, PMID 41299487). In 103 at-risk RA patients, a negative lung ultrasound plus normal DLCO had 100% negative predictive value (95% CI 82–100) against HRCT, while ultrasound alone had sensitivity 60.9% (Kouna 2026, PMID 42873714); ERS/EULAR and ACR/CHEST favor risk-based HRCT/PFT screening whereas ALAT admits ultrasound as an adjunct (Garcia-Salinas 2026, PMID 42888356). ERS/EULAR Table 1b explicitly recommends against replacing HRCT with PFTs and suggests against replacing it with lung ultrasound (Antoniou 2026, PMID 40907995). A two-step rule-out strategy therefore needs prospective safety and outcome testing before it can change that position. A prospective risk-based screening study should measure stage shift, treatment change, radiation, incidental findings and mortality.
OQ-8 — Which treatment best slows progressive RA-ILD?¶
Guideline options are conditional and rituximab/JAK evidence is mostly observational (Saavedra 2025, PMID 39822854; Krishna Boppana 2024, PMID 38933731; Narváez 2024, PMID 39270812). A 39-person single-arm tofacitinib RA-ILD phase II reported 92% any adverse event, three deaths, and 15/39 mortality or progressive pulmonary fibrosis, with FVC −72 mL (Rojas-Serrano 2026, PMID 42722091). A platform trial stratified by UIP/non-UIP and inflammatory/fibrotic progression is needed.
OQ-9 — When should antifibrotic and immunomodulatory therapy be combined in RA-ILD?¶
RA-ILD can contain inflammatory and fibrotic processes simultaneously, but combination sequencing lacks direct evidence (Kadura 2021, PMID 34168062; Saavedra 2025, PMID 39822854). A factorial or adaptive trial could test lung and joint outcomes together.
OQ-10 — How should difficult-to-treat RA be decomposed before drug switching?¶
A 23-study synthesis estimated EULAR-defined D2T prevalence at 11.7% (95% CI 9.5–14.3), but only 47.1% (33.3–61.4) had persistent inflammatory refractory disease; prevalence fell to 4.3% after ≥3 and 1.6% after ≥4 advanced-therapy failures (Xie 2026, PMID 41188120). The living Cochrane network meta-analysis of 19 RCTs (4,779 people) after b/tsDMARD failure found nine therapies with high-certainty ACR50 benefit versus placebo but no significant pairwise differences between active therapies, at low certainty (Thomas 2026, PMID 42440279); drug ranking therefore cannot resolve the switching decision. A 186-person retrospective cohort found higher 12-month retention when cycling to a second JAK inhibitor than switching to a biologic after first-JAK failure (HR 0.27, 95% CI 0.10–0.71) (Tsai 2026, PMID 42869207). A standardized pathway that separates inflammation, damage, nociplastic pain, adherence/access and comorbidity should be compared with usual mechanism cycling.
OQ-11 — Is triple conventional therapy noninferior to immediate biologic escalation in modern target-based care?¶
RACAT found triple therapy noninferior to etanercept–methotrexate for DAS28 change (upper 95% confidence limit 0.41 below a 0.6 margin), while TEAR found a small radiographic advantage for etanercept at two years (O'Dell 2013, PMID 23755969; Moreland 2012, PMID 22508468). A 29-trial meta-analysis of 13,387 methotrexate-naive people found higher DAS28 remission with initial b/tsDMARD plus methotrexate than methotrexate alone (6-month RR 1.62, 95% CI 1.39–1.89) (Russell 2026, PMID 42755040). That comparison is not a modern treat-to-target methotrexate-first sequence versus immediate biologic. A pragmatic trial should incorporate rapid methotrexate optimization, biosimilar pricing, speed, radiographs, adverse events, work and cost.
OQ-12 — What is the safest effective glucocorticoid bridge?¶
CareRA suggested faster functional improvement with a step-down bridge, but a 395-person placebo-controlled trial found no one-year radiographic advantage for high-dose bridging and no week-52 activity difference (Verschueren 2015, PMID 25889222; Krause 2022, PMID 35643951). In a 2,222-person Canadian early-RA cohort, glucocorticoid use at 12 months was 47% after early oral exposure versus 26% after parenteral exposure and 8% without early glucocorticoids; route was not randomized, so confounding by indication remains (Fernández-Codina 2026, PMID 42665531). Because infection risk is dose-dependent even at low sustained doses (George 2020, PMID 32956604), a contemporary trial should randomize route and fixed rapid tapers within methotrexate escalation using cumulative exposure and infection as co-primary safety measures.
OQ-13 — Can residual pain be assigned to an actionable mechanism?¶
Symptoms persist despite formal targets and may reflect inflammation, damage, neuropathy, fibromyalgia, sleep or mood (Michaud 2021, PMID 32619340; Buch 2021, PMID 33293696). A mechanism-stratified care trial should require objective inflammation assessment before randomization.
OQ-14 — Which intervention meaningfully reduces RA fatigue?¶
Fatigue is weakly tied to inflammatory activity and strongly linked with pain, sleep, mood, obesity and inactivity (Katz 2017, PMID 28386762; Pope 2020, PMID 32385141). A factorial trial could test activity, sleep and psychological components with a multidimensional fatigue endpoint.
OQ-15 — Does guideline-concordant comorbidity review reduce hard outcomes?¶
RA carries excess mortality, cardiovascular disease, infection and fracture risk (Lee 2024, PMID 38918258; Mehta 2019, PMID 31245055; Jin 2018, PMID 29546507). Cluster-randomize coordinated annual review against usual care and measure admissions, fractures, events and mortality.
Tier 2 — enabling science and longer-horizon questions¶
OQ-16 — Where does tolerance first break?¶
Lung, periodontal and gut mucosa each have mechanistic and associative evidence, but no universal origin (Lucchino 2019, PMID 31295951; Möller 2020, PMID 32582191). Longitudinal multi-mucosal sampling before ACPA emergence is required.
OQ-17 — Are ACPA-positive and ACPA-negative RA convergent endpoints of different diseases?¶
Genetic and synovial differences are reproducible, yet treatment algorithms remain shared (van Heemst 2014, PMID 24813459; Perera 2024, PMID 38727279). Multi-ancestry inception cohorts with tissue and treatment-response data are needed.
OQ-18 — Which autoantibody features are pathogenic rather than predictive?¶
ACPA titer, breadth and glycosylation change before onset, but causal components remain uncertain (Greenblatt 2020, PMID 32205569). Antigen-specific functional studies must connect human trajectories to tissue injury.
OQ-19 — Are fibroblast states stable therapeutic targets?¶
Single-cell data identify lining/sublining programs associated with inflammation and destruction, but causality and temporal stability are unresolved (Schonfeldova 2022, PMID 34559213; Komatsu 2022, PMID 35705856).
OQ-20 — What sustains remission after cytokine blockade?¶
Many patients relapse after withdrawal despite clinical remission (Tascilar 2021, PMID 38297524). Paired tissue before treatment, remission and taper could distinguish immune suppression from restored homeostasis.
OQ-21 — Can a blood multi-omic signature transport across ancestry and treatment contexts?¶
A 278-person longitudinal RA proteome cohort produced onset/activity/response signals, but external utility is unproven (He 2025, PMID 40691443). A 201-person early-RA cohort found that adding TNFA methylation to clinical data increased the methotrexate-remission model AUC from 0.699 to 0.760, but this is not an externally validated treatment-selection test (Dalix 2026, PMID 42664067). Locked models need multinational prospective validation across ancestry and treatment contexts.
OQ-22 — Does the microbiome cause, amplify or merely reflect RA?¶
Case-control signatures are sensitive to diet, geography and DMARD exposure (Möller 2020, PMID 32582191; Qi 2025, PMID 40825448). Pre-treatment longitudinal sampling and perturbation studies are prerequisites.
OQ-23 — Can smoking cessation after ACPA appearance prevent arthritis?¶
Smoking is a strong ACPA-positive risk factor, but cessation-specific prevention effects are not established (Lucchino 2019, PMID 31295951). A randomized cessation-support trial can test biomarker evolution even if arthritis onset requires larger follow-up.
OQ-24 — What explains organ-specific disease in RA?¶
Only subsets develop ILD, vasculitis, scleritis or Felty syndrome despite shared systemic autoimmunity (Kadura 2021, PMID 34168062; Mertz 2023, PMID 37468085; Turk 2021, PMID 32358156). Cross-organ immune atlases should compare tissue clonotypes and stromal programs.
OQ-25 — Can immune reset produce drug-free remission in refractory RA?¶
Six compassionate blinatumomab cases showed rapid improvement, while CAR-T programs are emerging (Bucci 2024, PMID 38671240). A 15-person low-dose series reached week-12 CDAI LDA in 9/15 but 14/15 later flared, with 7/15 then responding to previously failed DMARDs (Bucci 2026, PMID 42595657). In the first RA CD19 CAR-T phase 1 (COMPARE, n=6, ACPA-positive refractory disease, all DMARDs stopped), CRS occurred in 6/6 (grade 1–2), with no ICANS; B cells were depleted in blood and tissue and IgM-RF seroconverted in 5/6, yet only 3/6 reached DAS28-CRP remission and the median DAS28-CRP reduction was 34% at 36–52 weeks (Albach 2026, PMID 42661098). Deep serologic reset without uniform clinical remission is itself informative: it suggests B-cell-independent drivers persist in some refractory joints. The phase 2 stage randomizes CAR-T against rituximab (NCT06475495), and an in vivo CD19 CAR-T phase 1 including RA was first posted on 2026-10-14 (NCT07873931). Controlled studies must establish durability, infection risk, neurotoxicity and whether tissue pathology is eliminated. NCT07796516 is recruiting telitacicept in refractory RA.
OQ-26 — How much structural repair is biologically possible?¶
Inflammation activates osteoclasts and suppresses osteoblast repair; erosions rarely normalize despite remission (Komatsu 2022, PMID 35705856). Imaging and tissue studies should distinguish arrested damage from true repair.
OQ-27 — Why do validated targets hit a response ceiling?¶
Roughly 40% show poor response to an individual biologic and 5–20% may be refractory across mechanisms (Humby 2021, PMID 33485455; Bhamidipati 2022, PMID 35248489). Explanations include wrong target, drug delivery, tissue redundancy, irreversible damage and outcome misclassification.
OQ-28 — Can trial outcomes represent participation and treatment burden?¶
Patient-reported-outcome literature establishes pain, fatigue and function, but dosing/monitoring/access burden is less standardized (van Tuyl 2016, PMID 27133486). A core treatment-burden set should be co-developed internationally.
OQ-29 — How should global RA burden be measured where primary data are sparse?¶
GBD projects 31.7 million cases by 2050, but modelled estimates depend on sparse inputs and case definitions (GBD 2021 RA Collaborators 2023, PMID 37795020). A GBD 2023 analysis estimated 12.6 million cases across 129 LMICs and a 4.1-fold prevalence difference between upper-middle- and low-income countries, a gradient that may mix true disease frequency with diagnostic capacity and data availability (Wang 2026, PMID 42495527). Validated low-cost case-finding linked to rheumatology review is needed to distinguish burden from ascertainment.
OQ-30 — Does risk-based RA-ILD screening improve outcomes rather than only detection?¶
ACR/CHEST conditionally recommends PFTs and HRCT for at-risk patients, while pooled RA-ILD prevalence is 18.7% with I²=96.4%, showing how strongly ascertainment changes yield (Johnson 2024, PMID 38973714; Wang 2024, PMID 38547537). NCT07849673 will test whether EHR-embedded decision support increases screening in RA and SSc; it is a process trial, not an outcomes trial. A randomized or stepped-wedge program should test stage shift, treatment change, radiation/incidental findings, FVC decline, hospitalization and mortality.
OQ-31 — Which part of the nintedanib effect is additive to optimized immunomodulation in progressive RA-ILD?¶
In the 89-person INBUILD RA-ILD subgroup, nintedanib slowed FVC decline by 116.7 mL/year (95% CI 7.4–226.1), but background DMARDs were not assigned and diarrhea occurred in 61.9% (Matteson 2023, PMID 37209188). A factorial platform should separate joint/inflammatory control from antifibrotic effect and measure discontinuation and quality of life.
OQ-32 — Can remote monitoring reduce visits without widening digital inequity?¶
Controlled ePROM studies suggest a small disease-activity benefit (SMD −0.15, 95% CI −0.27 to −0.03), but most were high risk of bias; AEGORA did not improve self-efficacy and retained only 43% of app users at week 16 (Arumalla 2023, PMID 37204273; Doumen 2025, PMID 39576683). NCT07857993 (DETECTRA) will test remote PROs plus self-imaging for earlier flare detection. A service trial should stratify digital access, literacy and language and test missed flares, visit burden and equity.
OQ-33 — Can vagus-nerve modulation deliver durable benefit with acceptable procedural risk?¶
RESET-RA achieved ACR20 in 35.2% with active versus 24.2% with sham stimulation at three months (p=0.0209), with related serious events confined to the perioperative period, but open-label follow-up cannot establish durability against sham (Tesser 2026, PMID 41429981). A longer blinded comparison should include device revision, infection, patient burden and cost.
OQ-34 — Which preference-sensitive decisions materially improve adherence and outcomes?¶
Across 36 studies, many patients weighted benefit above route or adverse effects, but preference patterns varied and discrete-choice methods were heterogeneous (Durand 2020, PMID 30988125; Zartab 2021, PMID 34447905). A trial of preference-concordant versus standard sequencing should prespecify persistence, disease control, regret and equity.
Dots not yet connected¶
| # | Dot A | Dot B | The missing junction | Powers |
|---|---|---|---|---|
| D-1 | In 49 people at risk of rheumatoid arthritis, sputum IgA ACPA correlated with the share of neutrophils forming citrullinated-histone NETs (r = 0.49, P = 0.002) and with NET remnants (r = 0.70, P < 0.001) (PMID 34369110) | In 18 ACPA-positive people with clinically suspect arthralgia and no joint swelling, synovial fibroblast-activation PET uptake predicted clinical rheumatoid arthritis (5/18; HR 4.58, 95% CI 1.38–15.22) over a median 42 weeks (PMID 41314671) | PubMed and Europe PMC searches found no study testing whether a sputum or lung ACPA trajectory determines that pre-arthritis fibroblast state. | Prevention + tissue precision |
| D-2 | In ARIAA, arthritis developed by 6 months in 4/49 (8%) abatacept recipients versus 17/49 (35%) placebo recipients with ACPA-positive arthralgia and MRI inflammation (PMID 38364841) | In established RA, the 164-person R4RA biopsy analysis predicted rituximab and tocilizumab response with cross-validated AUCs of 0.74 and 0.68 (PMID 35589854) | No PubMed-indexed prevention study has tested whether a pre-arthritis synovial tissue signature selects an interceptive drug; R4RA sampled joints after established disease and TNF-inhibitor failure. | Prevention + precision trials |
| D-6 | At least one in six people with RA has severe fatigue in the reviewed evidence (PMID 32385141) | Across 23 studies, only 47.1% (95% CI 33.3–61.4) of difficult-to-treat RA cases had persistent inflammatory refractory disease (PMID 41188120) | In ARCTIC, clinically relevant fatigue fell from 69% to 38% at 24 months, and low swollen-joint and power-Doppler scores plus a high patient global assessment predicted later fatigue (PMID 34389605). PubMed and Europe PMC searches found no trial that assigns fatigue care by an objective inflammatory versus non-inflammatory difficult-to-treat phenotype. | Patient outcomes + imaging |
| D-7 | In the Medicare RA cohort, one-year hospitalized infection incidence was 8.6% without glucocorticoids versus 11.0% (95% CI 10.6–11.5) at ≤5 mg/day (PMID 32956604) | In CAMERA, monthly protocolized methotrexate adjustment yielded at least one remission period in 50% versus 37% with three-monthly conventional care (n=299; PMID 17519278) | No PubMed-indexed trial tested whether faster methotrexate escalation reduces cumulative steroid exposure and hospitalized infection while preserving flare control. | Strategy + safety |
| D-8 | Thirteen observational studies pooled a periodontitis–RA association of OR 1.69 (95% CI 1.31–2.17), which does not establish causality (PMID 32593704) | In the ACPA-positive, MRI-enriched ARIAA prevention trial, arthritis developed by 6 months in 8% on abatacept versus 35% on placebo (PMID 38364841) | No PubMed-indexed trial randomized periodontal treatment in an at-risk pre-arthritis cohort with first-arthritis onset as an endpoint. | Dentistry + prevention |
| D-14 | In ORAL Surveillance, people aged ≥50 with active rheumatoid arthritis despite methotrexate and at least one cardiovascular risk factor had more major cardiovascular events (HR 1.33, 95% CI 0.91–1.94) and cancers (HR 1.48, 95% CI 1.04–2.09) on combined tofacitinib doses than on a TNF inhibitor; noninferiority against a margin of 1.8 was not shown (PMID 35081280) | In the 164-person R4RA biopsy trial, nested cross-validation predicted multidrug resistance with an AUC of 0.69 (PMID 35589854) | No PubMed-indexed study was found that tests whether a synovial pathotype, or that resistance model, predicts serious infection, cancer, or major cardiovascular events. | Precision + safety |
| D-16 | Across eight controlled inflammatory-arthritis ePROM studies (4,473 people; seven studies included RA), remote monitoring reduced disease-activity scores by SMD −0.15 (95% CI −0.27 to −0.03); most studies had high risk of bias (PMID 37204273) | In PRESERVE, patient ratings exceeded physician global ratings by ≥2 points in 25.5% at baseline and 24.8% after 36 weeks (PMID 27209012) | No PubMed-indexed intervention study has tested whether remotely detected discordance should trigger objective synovitis assessment and a different care pathway for pain or fatigue. | Patient outcomes + services |
| D-18 | In the 242-person RESET-RA trial, three-month ACR20 response was 35.2% with active vagus-nerve stimulation versus 24.2% with sham (PMID 41429981) | A synovial atlas of more than 314,000 cells from 79 donors resolved six cell-type abundance phenotypes (PMID 37938773) | No PubMed-indexed response study paired vagus-nerve stimulation with synovial cell-state profiling to test whether a tissue phenotype predicts benefit. | Bioelectronics + tissue biology |