[ad_1]
September 08, 2025
8 min read
AI has within the last few years burgeoned into a ubiquitous presence in physician offices and medical facilities across multiple specialties, including rheumatology.
And while its most common uses still appear to be in coding, letter writing and transcribing, AI has also been gaining ground — and utility — at the point of care.
“AI is no longer a distant concept,” Leonard H. Calabrese, DO, professor of medicine at the Cleveland Clinic Lerner College of Medicine of Case Western Reserve University, and RJ Fasenmyer chair of clinical immunology at the Cleveland Clinic, told Healio. “From ambient listening and automated scribing to machine learning algorithms for disease detection, AI-assisted imaging and tools that synthesize massive medical records, these technologies are now either embedded in daily practice or rapidly approaching widespread adoption.”
Speaking at the recent Advances in Rheumatology and Technology Symposium (ARTS), a 2-day meeting in Chicago covering the intersection of the specialty and emerging innovations, Calabrese, who is also chief medical editor for Healio Rheumatology, said point-of-care AI can help rheumatologists untangle difficult diagnoses.
According to Calabrese, the task of diagnosing and managing multisystem disorders — “cases that can be as challenging as they are rare” — is often made even more difficult by the “traumatic growth” of medical literature. He estimated that the number of entries in this vast collection, including clinical trials, studies and reviews, doubles every 5 years.
“AI offers a way to keep evidence at our fingertips, complementing — and in some cases replacing — the time-intensive process of combing through textbooks, online publications, PubMed searches, or even well-established tools like UpToDate, which itself is now incorporating AI-driven features,” Calabrese said.
He shared the stage at ARTS, in a session on polymyalgia rheumatica (PMR) and vasculitis, with Anisha Dua, MD, MPH, director of the Vasculitis Center at the Northwestern University Feinberg School of Medicine, and Michael Putman, MD, MSci, medical director of the Vasculitis Program at the Medical College of Wisconsin, who moderated the discussion.
According to Dua, the role of emerging imaging modalities — such as radiomics, which uses AI to extract quantitative metrics from medical images, as well as enhanced precision and new, more specific tracers — is “rapidly evolving.” These modalities are currently being used in the assessment of large vessel vasculitis (LVV) and PMR, she added.
“We are significantly enhancing our ability to diagnose and monitor these patients through these technologic advancements,” Dua told Healio. “This is particularly true in LVV, where there are no great serologic biomarkers and tissue pathology is not regularly attainable or accessible.”
Meanwhile, Putman suggested that although large language models such as ChatGPT can be “amazing” at certain jobs, particularly coding, he rarely uses them in clinical settings. Specifically, they can be useful in providing diagnostic possibilities that a human clinician can then analyze and use to make a determination, rather than in making diagnoses on their own, he said.
“I think AI is very helpful and it’s getting better all the time,” Putman told Healio. “At this point, it’s clear to me that it’s not ready to diagnose people with things. It’s just not quite there. However, it’s very useful in helping you think of the range of possibilities.”
Overcoming ‘cognitive bias’
According to Calabrese, rheumatologists — “like all clinicians” — are susceptible to cognitive errors and biases that contribute to diagnostic mistakes. He added that he often turns to AI to help him aggregate and organize evidence and the ever-expanding sea of medical literature and data.
Although he acknowledged the growing presence of multiple names in that space, Calabrese specifically highlighted OpenEvidence, which earlier this year announced partnerships with the New England Journal of Medicine and the JAMA Network to provide their published content to inform the platform’s answers to medical questions.
“OpenEvidence is a free tool for any practitioner with a National Provider Identifier number and is now the fastest-growing AI platform in medicine, with subscriptions from more than one-third of practicing physicians in the United States,” Calabrese said. “Accessible both online and via a mobile app, OpenEvidence aggregates, synthesizes and visualizes clinically relevant evidence, and now partners with high-impact journals including the New England Journal of Medicine and the JAMA Network, providing full-text access and graphical summaries.”
According to Dua — who said she uses both OpenEvidence and ChatGPT — these data aggregation and large language models can help write letters fighting insurance denials, as well as compile large swathes of data and present them to clinicians in an easily digestible package.
However, she cautioned that such platforms require “the right inputs” and human verification to correct any “hallucinations” that appear in the results or prose. Additionally, they are not designed to replace person-to-person interactions with patients.
“I think AI can access a lot of relevant data and compile it into understandable information that will be able to help clinicians at the point of care in diagnostically challenging cases,” she said. “However, it will take the right inputs and of course the human interaction with the patient to drive the management plan and assessment.
“I do use AI in my practice, sometimes for point-of-care, but often for writing letters to fight denials by insurance companies for drug or imaging study coverage,” Dua added. “I do find it helpful, but always review whatever content is created for accuracy and add patient-specific information separately.”
Dua added that AI in imaging can also improve workflow for clinicians. These models, along with technologic advancements in imaging platforms and techniques, can reduce the time needed for, and increase the reproducibility of interpreting, PET scans by limiting the need for manual segmentation of the aorta and its branches, she said.
“There are new AI models that are also being developed to help with interpretation of Ultrasound images in giant cell arteritis,” Dua said. “All of these are exciting and evolving areas in the area of LVV assessment and will hopefully lead to improved outcomes for our patients.”
Meanwhile, Putman, who described himself as an early adopter of ChatGPT, and says he is among the top 10% of the platform’s users among his colleagues, uses AI “constantly” for non-clinical work.
“It’s amazing at coding,” he said. “I can’t believe how good it is at coding. I’d say that I’m probably 10 times faster at coding than I was 2 years ago. It’s just unbelievable what it can do in that realm.”
Putman also praised the platform’s ability to write and provide first-draft outlines for meeting presentations.
“I never use ChatGPT to write things for me, but sometimes I’ll use a turn of phrase, like ‘What do you think about this sentence here?’ It’s incredibly good at that kind of thing,” he said. “If you’re giving talks, it’s really good at giving the skeleton of a talk.”
Making the case
To demonstrate the potential of AI at the point of care, Calabrese presented a case of suspected IgG4-related disease. He described using AI as if “conversing with a trusted colleague,” presenting key history, examination and initial test findings, then using the platform to generate and triage the most likely differential diagnoses.
According to Calabrese, one of the strengths of the program — in this case, OpenEvidence — is that it suggests follow-up questions, which he said often present new insights. Calabrese also praised its ability to revisit and review prior searches and conversations, which are automatically archived. He added that he has used the platform more than 2,000 times since adopting it in July 2024, “not only for patient cases but also for tangential learning on immunopathologic pathways, emerging therapies, and evolving literature.”
In his case presentation, the platform refined the differential diagnosis, underscored essential “must-rule-outs,” and surfaced high-impact references for deeper evaluation, Calabrese said.
“In complex medical cases, diagnosis arises from the interplay of knowledge and skill,” he said. “Master clinicians rely not only on experience but also on a keen sense of nuance, sifting through clinical details, contextual clues and the patient’s story in ways AI cannot replicate.
“When uncertainty arises, we traditionally turn to journals, textbooks, online resources, and the counsel of trusted colleagues,” Calabrese added. “These approaches, though invaluable, are often time-consuming. Point-of-care AI tools such as OpenEvidence now allow rapid access to this breadth of information.”
That said, Calabrese additionally stressed that AI “should never replace” physician judgment or assume the role of final decision-maker.
“Instead, it should serve as a powerful partner, offering evidence and references of peer-reviewed articles — as OpenEvidence does — within minutes that we can then critically appraise,” he said.
Calabrese added that many other AI programs he has tested are not as transparent as OpenEvidence regarding their sources, and often provide data that may not be peer reviewed.
“Above all, whether information comes from AI or a renowned authority, it must ultimately align with our own reason, judgment, and common sense,” he said.
Putman additionally offered a complex case, involving a patient with high alkaline phosphatase levels, in which he turned to ChatGPT for suggestions.
“I had a complicated case recently where I was trying to figure out why someone had a really high alkaline phosphatase level, and I asked it a bunch of questions about why the alkaline phosphatase would be so high, and what the possible causes were,” he said. “The platform kind of does the same thing that WebMD did, where it gives you all the possible imaginable things, including cancers.
“I think of it as a really useful tool, but as far as human replacement level, I don’t think it’s anywhere close yet,” he added.
That said, such tools are growing more refined and are constantly improving with time, according to Putman.
“In a long enough time frame, I think it’s essentially guaranteed,” he said. “I think you’d be hard-pressed to find anyone who doesn’t think that some form of AI will be able to replicate human-level reasoning and thought. I’d be astonished if it wasn’t capable of that eventually. However, I think the ‘eventually’ is further out than people realize.”
‘The beginning of the beginning’
According to Calabrese, the sum of all advances seen in AI up to this point merely represents the technology in its infancy.
“We are not at the beginning of AI,” he said. “We are at the beginning of the beginning of AI.”
With multiple new platforms emerging, Calabrese predicted their long-term impact will ultimately be determined by the “quality, transparency and relevance of their content.”
Putman, meanwhile, predicted a trajectory for AI not unlike the early days of the commercial internet — an initially overvalued bubble followed by more stabilized long-term growth.
“I think we’re in a bubble right now,” he said. “You know the original dot-com bubble? In 2000, everything crashed and everyone said, ‘Oh my gosh, we built all these Pets.com companies and they wound up failing,’ and they were all grossly overvalued and it was a big disaster.
“It was true that we were in a bubble, but then 10 years later it was all correct — we really do everything online,” he added. “Amazon lost a lot of its value during the dot-com crash, but in retrospect it was grossly undervalued on a longer-term horizon. It’s an enormous, incredibly profitable company now. I think AI is similar, where people are probably overvaluing it short-term and undervaluing it long-term.”
Whatever the short- or long-term outlook of AI, Dua said she hopes clinicians learn to use the technology to enhance patient care, rather than become reliant on it or attempt to use it to replace iterative thinking and clinical judgement.
“I think it offers a lot of value across multiple areas of medicine including education, patient care and access,” she said. “It holds a lot of promise in streamlining and optimizing medical processes. I do think we are on a new horizon, and it will be very exciting to shape and discover how AI interfaces with the medical field to open up new possibilities for providers and patients.”
Reference:
Calabrese LH, Dua A, Putman M. PMR/Vasculitis. Presented at: Advances in Rheumatology and Technology Symposium; Sept. 5-6, 2025; Chicago.
For more information
Leonard H. Calabrese, DO, can be reached at calabrl@ccf.org.
Anisha Dua MD, MPH, can be reached at anisha.dua@northwestern.edu.
Michael Putman, MD, MSci, can be reached at mputman@mcw.edu.
Sources/Disclosures
Collapse
Healio interviews.
Disclosures:
Calabrese reports being a medical advisor for OpenEvidence, as well as professional relationships with AbbVie, AstraZeneca, Bristol Myers Squibb, Galvani, Genentech, GlaxoSmithKline, Janssen, Novartis, Regeneron, Sanofi and UCB. Dua reports receiving consulting fees from AbbVie, Amgen, AstraZeneca, GlaxoSmithKline, Sanofi, Novartis and Zenas; and grant support from Novartis, Amgen, Rheumatology research foundation. Putman reports receiving research funding from AbbVie and Amgen, research and other compensation from AstraZeneca and Novartis, and being on the advisory or review panel for GlaxoSmithKline.
[ad_2]
Source link
