CHICAGO—ACR Convergence 2025 featured a diverse and influential body of research advancing the field of rheumatology. Abstracts and presentations with a focus on rheumatoid arthritis (RA) spanned the disease continuum. This synopsis highlights key areas from the meeting, including strategies to delay or prevent RA, emerging mechanisms of action in RA treatment, and the growing role of artificial intelligence (AI) and digital interventions in advancing RA research and clinical care.
Can We Delay or Prevent RA?
Abstracts 1678, 1674 & 10341-3
Patients and clinicians have continued interest in preventing RA in high-risk individuals, such as those who are ACPA+, RF+ or with symptoms suggestive of RA but without signs of synovitis. Cope et al. reported results from the extension to the APPIPRA trial in the ALTO study to further examine whether one year of subcutaneous abatacept prevents or delays the development of RA through five years and beyond.
In the same vein, Deane et al. reported extended results from the phase 2 STOP-RA trial, evaluating hydroxychloroquine (HCQ) to prevent RA. Patients were anti-CCP3 positive and had no history or exam-based evidence of inflammatory arthritis at baseline. They were identified as patients who presented to recruiting clinics; were recruited from first-degree relatives of those with established RA; or were recruited at health fairs, biobanks and blood donation sites.
A total of 144 patients were randomized (1:1) to HCQ vs. placebo for 12 months, with HCQ dose at ≤6.5 mg/kg, based on ideal body weight. The trial reported no difference between HCQ and placebo in the proportion developing clinical RA (34% HCQ vs. 39% placebo (P=0.52) at 36 months. Importantly, there were no baseline factors (e.g., RF-IgM positivity) that suggested any benefit to HCQ over placebo.
This trial, while disappointing given that HCQ has a relatively benign side effect profile, is nevertheless informative to clinicians to confirm that HCQ does not modulate the required biologic target(s) for RA prevention. It also illustrates the importance of selecting the highest risk patients for future RA prevention trials, in as much as only about one-third developed RA over three years in the STOP-RA trial, whereas in other RA prevention trials, as many as twice that proportion eventually develop RA.
While people at high risk for RA who haven’t yet developed the disease may not be willing to take a biologic drug for a year or more to prevent disease onset, a common question is whether such patients can implement behavioral or lifestyle changes to possibly impact their risk. A meta-analysis of healthy dietary pattern trials and observational studies was reported with the goal to see if certain diets may confer a lower risk of incident RA. The researchers found that people observing certain diets, including an anti-inflammatory diet (odds ratio [OR] 0.56, 95% CI 0.31–0.99) and a Mediterranean diet, had a lower risk of developing RA (OR 0.88, 95% CI 0.78–0.99).
Given the low risk of a dietary intervention to prevent RA, this new evidence from the meta-analysis is likely helpful for providers when counseling patients at high risk.
New RA Mechanisms of Action for Therapeutic Intervention: Vagal Nerve Stimulation
Abstracts 2278, 2614 & 16754-6
There have been few, true, new advances in treatments for RA in the past decade. Janus kinase (JAK) inhibitors were first approved for RA in 2012 with the advent of tofacitinib, and although new JAK inhibitors (i.e., baricibinib and upadactinib) and other in-class medications, such as sarilumab (an interleukin-6 receptor inhibitor [IL-6Ri]), have been approved since, few therapeutic advances for RA offer new mechanisms of action.
Tesser et al. provided 12-month results from the RESET-RA trial implantable using a vagal nerve stimulation (VNS) device. Results for the primary outcome were significant at 12 weeks (ACR20 35.2% [VNS] vs. 24.4% in sham control-treated patients, P=0.0209), and response rates continued to improve over time.
By six months, 52.1% of patients achieved an ACR20 response. A similarly high proportion (49.3%) attained DAS28CRP low disease activity, and 47.4% attained CDAI low disease activity or remission. Only 24.8% of patients added a biologic or targeted synthetic DMARD to VNS treatment over 12 months. There were no serious infections in the first three months, and the serious adverse event rate in the treated patients (3.3%) was numerically lower than in sham-control patients (4.2%).
A few key subgroup analyses were also reported at Convergence Convergence, in separate presentations. In the subgroup of patients with exposure to only one prior tumor necrosis factor (TNF) inhibitor, response at three months was even better than in the overall trial: VNS 48.0% vs. sham control 21.2%, P=0.02. By 12 months, 55.7% of patients with prior exposure to only TNF inhibitors attained an ACR20 response, and 36.4% achieved an ACR50 response. And in the subgroup of patients enriched for progression of radiographic damage, the associated MRI study that evaluated OMERACT-validated imaging using gadolinium-enhanced MRI demonstrated a lower proportion of erosion progression at three months (18.9%) compared with sham (37.8%, P=0.015).
The implications of this trial are that VNS offers patients a new therapeutic option for RA that appears both effective and safe, and harnesses the body’s own neuroimmune pathways. Based on the mechanism of action and safety profile of VNS as observed in this trial, it seems likely that one could add a biologic or JAK inhibitor, if needed, without adverse consequences (e.g. increased rate of serious infections).
Compared with current options in RA that require patients to switch, rather than add, advanced therapies after failure of conventional synthetic DMARDs, the potential to combine multiple advanced mechanisms of action may raise the bar for therapeutic response, allowing more patients with RA to attain remission.
From a research perspective, this trial may set a precedent to use validated MRI outcomes to rapidly evaluate changes in erosions and synovitis using OMERACT-validated outcome scores measured as early as three months. This approach lessens the field’s dependence on long and costly X-ray studies that require huge numbers of patients to show small differences in X-ray outcomes and may provide an important advance in future RA trial design.
Digital Psychological Support: There When Your Rheumatologist Isn’t
Abstract 16607
The pressures of a busy office practice leave almost all clinicians wishing they had more time to interact with their patients. And despite the substantial array of excellent medication choices available for patients with RA and other inflammatory arthritides, the need to help patients manage psychological distress, chronic pain (not necessarily related to inflammation) and social support is time consuming. Knitza et al., a European research team, randomized patients to receive digital support via a mobile app that taught the principles of cognitive behavioral therapy (CBT) with a comparator group.
Over the 12 weeks of the study, psychological distress, measured by the Hospital Anxiety and Depression Scale (HADS), stress and anxiety all substantially fell compared with the control group. Most patients were satisfied with the intervention.
Two features of this approach deserve mention. The first is that it is one of the still-uncommon examples of a decentralized trial in rheumatology. Decentralized trials reflect a design that typically requires little or nothing of a site investigator or her staff, because most or all enrollment activities, recruitment, intervention and data collection are done remotely (i.e., not at the site, nor by its staff).
The second is that digital tools can offer unparalleled efficiencies to deliver services and care directly to patients, without requiring the precious time of the clinician at in-person visits.
It is likely that as a field, rheumatology will continue to benefit from advances in digital medicine to remotely deploy services not typically offered by rheumatology providers (e.g., CBT, pain management, physical or occupational therapy, dietary interventions), advancing the care of patients in out-of-office settings.
Combating Disparities in RA Management
Abstracts 0478 & 16568,9
In the U.S., older patients with late-onset RA commonly have Medicare insurance coverage, which provides access to biologics and targeted therapies for many patients under the Part B buy-and-bill benefit. But do all patients with RA have equal access to these treatments?
Ara et al. conducted a large-scale analysis using a 20% random sample of Medicare data from 2016–20 to evaluate whether older and non-white patients, and women had equal access to these treatments.
Focusing on all RA specialty drugs, race and ethnicity were not associated with receipt of these treatments, but women were 13% less likely to receive these treatments, and those receiving a low-income subsidy (a proxy for lower socioeconomic status) were 15% less likely to receive specialty medications. Turning to Part B-covered drugs (e.g., infliximab, IV abatacept and provider-administered certolizumab), patients who were Black or Hispanic were 22% and 25%, respectively, less likely to receive Part B RA medications. Moreover, each additional year of age was associated with a 5–6% lower odds of using a specialty or Part B drug for RA.
What do we make of these findings? Do older patients and those of certain racial or ethnic backgrounds truly have less severe or active RA and not need these therapies due to biologic differences in their disease? That explanation seems unlikely.
Rather, access to care, provider trust, evaluation of RA medications’ risk-benefit profiles and ageism (i.e., treating older patients’ RA less aggressively, perhaps due to a higher prevalence of age-related comorbidities or other factors, which may or may not be medically justified) are more likely explanations. Indeed, in a separate presentation, Lee et al. used data from the patient registry FORWARD, with date on 873 patients with RA (2001-2019) and found the risk for serious infections associated with non-TNF inhibitor biologics was minimally different in older patients with RA (at least through age 70) vs. younger patients with RA after controlling for disability, glucocorticoid use and prior infections, suggesting that differential under-prescribing of biologics based solely on older age may not be warranted.
What is unknown from the analysis of the Medicare analysis is whether patients were offered these treatments by their provider and declined, or for other reasons (e.g., costly co-insurance) were unable to access them. What is clear is that equal and affordable access to care, especially for expensive rheumatology treatments, is a laudable goal that we have yet to achieve.
The closure of the Medicare part D donut hole (i.e., coverage gap) in 2025 that instituted a hard cap for all out-of-pocket medications ($2,100 for 2026) that allows costs to be spread out over the year, and government-negotiated drug prices for certain immunologic therapies (e.g., etanercept, nintedanib, ustekinumab, apremilast) in 2026 and beyond may improve access to care. However, whether these policies reduce or eliminate disparities remains to be seen.
Harnessing Big Data in RA
Abstracts 1657, 0448 & 131010-12
Effective use of AI in rheumatology continues to be an exciting topic and was the source of numerous presentations, workshops and symposia at ACR Convergence 2025, including the session The Great AI Debate: Unlocking Potential or Unleashing Chaos? However, effective use of AI will be hampered if the quality of the data that it uses is poor. Numerous endeavors continue to apply AI and machine learning approaches within the rheumatology field. As one example presented at the meeting, researchers were able to identify reasons for switching TNF inihibitor treatments using GPT4 and other large language models (LLMs).
On average, GPT4 seemed to outperform the other LLMs and was able to identify reasons for stopping/switching with high accuracy (micro-F1 score 0.83, with 1.0 indicating perfect discrimination). As expected, the most common reasons for switching were lack of effectiveness, adverse events and insurance/cost.
Other abstracts reporting results from AI/LLMs found that RA flares were able to be accurately classified (95.6%) in hospitalized patients, and could accurately extract disease activity measures from unstructured physician notes.
The implications from this body of work suggest that in RA, and likely for other rheumatic diseases, large volumes of unstructured data (e.g., physician notes or hospital discharge summaries) can be automatically parsed and curated to extract important clinical information with high accuracy, without time-consuming manual medical record review.
Why does this matter? These advances are important because there are numerous applications of AI that require high-quality data for them to be useful, and AI itself appears an effective tool to generate high-quality data from narrative information that is part of a patient’s routine care (e.g., from the EHR).
A few use cases for AI include research topics, such as: 1) phenotyping patients (e.g., finding RA for clinical trial screening or finding subgroups of RA patients, such as those with interstitial lung disease); 2) gathering from unstructured data the information regarding patient’s treatment response and reasons for stopping/switching to build AI-based clinical and biomarker-based models to predict response to a new RA medication; and clinical topics, such as: 3) triaging and drafting an automated response to patients’ queries to their provider submitted via a patient portal; 4) extracting the necessary information required to populate prior authorization requests and letters of medical necessity; and 5) building clinical decision-support modules to provide evidence-based suggestions to providers.
The applications for AI are endless, but all benefit from high-quality data. AI itself offers a useful toolset to extract the required data from routine data sources, such as rheumatologists’ EHR systems.
Dr. Curtis
Jeffrey Curtis, MD, MS, MPH, is the Marguerite Jones Harbert–Gene Ball Endowed Professor of Medicine, Division of Clinical Immunology & Rheumatology, University of Alabama at Birmingham. A rheumatologist and epidemiologist, Dr. Curtis has research interests in rheumatoid arthritis, psoriatic arthritis, ankylosing spondylitis and osteoporosis, as well as methodologic interests in real-world evidence, causal inference, machine learning and artificial intelligence.
References
- Cope A, Vasconcelos J, Jasenecova M, et al. Abatacept in individuals at risk of developing rheumatoid arthritis: Results from the Arthritis Prevention in the Pre-clinical Phase of RA with Abatacept Long Term Outcomes (ALTO) Study [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Deane K, Striebich C, Feser M, et al. A phase 2, randomized, placebo-controlled trial of hydroxychloroquine in individuals at-risk for future rheumatoid arthritis [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Joerns E, Sparks J, Chelf C, et al. Healthy dietary patterns and risk of rheumatoid arthritis: A systematic review and meta-analysis [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Tesser J, Crowley A, Box J, et al. Neuroimmune modulation for the treatment of rheumatoid arthritis: Results at 12 months from a randomized, sham-controlled, double-blind study [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Peterfy, MD, PhD C, Tesser J, Levine Y, et al. Impact of vagus nerve-mediated neuroimmune modulation on structural joint damage using Gd-MRI RAMRIS imaging in biologic-experienced patients with rheumatoid arthritis [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Valenzuela G, Box J, Crowley A, et al. Neuroimmune modulation in patients with active rheumatoid arthritis with an inadequate response to TNF inhibitors (TNFi) [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Knitza J, Kraus J, Krusche M, et al. Digital psychological support for inflammatory rheumatic diseases: A randomized clinical trial [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Ara a, FitzGerald J, Ettner S. Racial and ethnic disparities in DMARD use and in Medicare Part B-covered options among Medicare beneficiaries with late-onset rheumatoid arthritis [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Lee J, Pedro S, Michaud K. Risk of serious infection associated with non-TNF biologic initiation after anti-TNF use in older adults with rheumatoid arthritis [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Miao B, Binvignat M, Garcia-Agundez A, et al. Extracting TNF inhibitor switching reasons and trajectories from real-world data using large language models [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Koulas I, Tsaftaridis N, Gkionis M, et al. Implementing artificial intelligence to identify rheumatoid arthritis flares using electronic medical records processed with privacy-preserving large language models: A pilot study [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).
- Park E, Kamdar I, Weisberg R, et al. Naturalized language processing based extraction of rheumatoid arthritis disease activity measures from the electronic health record [abstract]. Arthritis Rheumatol. 2025;77(suppl 9).