Sessions Will Explore Imminent Applications and Future Landscape of AI Tools in Rheumatology Clinical Care, Education, and Research

The past few years have marked the entrance of artificial intelligence (AI) into public consciousness and its expanding presence across domains of healthcare — administrative tasks, clinical education, risk prediction, and image analysis. AI innovations are at the door or just over the horizon in the field of rheumatology as well.

During ACR Convergence 2024, four sessions will provide attendees with exceptional opportunities to gain expert-led insights on how AI is percolating into rheumatology research and practice. All sessions will be available on demand within 48 hours for registered ACR Convergence 2024 participants.

AI and AI-Assisted Tools

In Harnessing the Power of AI in Rheumatology Without Getting Burned, on Saturday, Nov. 16, from 9:30–11 a.m. ET in Ballroom A of the Walter E. Washington Convention Center, John Isaacs, PhD, Professor of Clinical Rheumatology at University of Newcastle, United Kingdom, Bella Mehta, MD, Assistant Professor and Attending Rheumatologist at the Hospital for Special Surgery, Weill Cornell Medicine, and Amanda E. Nelson, MD, MSCR, Professor of Medicine and Adjunct Professor of Epidemiology at University of North Carolina at Chapel Hill, will explore the basics of AI and some AI-assisted tools of relevance to rheumatologists.

 “We are using AI in various aspects of our lives. Mainstream medicine has been slower to adapt compared to any other field, but AI in medicine is now where electronic medical records (EMRs) used to be 15–20 years ago,” Dr. Mehta said.

Dr. Mehta emphasized that AI algorithms are not going to replace rheumatologists now or in the foreseeable future; however, using AI to help patients and improve clinic workflows is imminent.

She cited the example of how AI technologies, especially large language models (LLMs), are being integrated into EPIC, one of the most popular EMR systems at present. EPIC has incorporated AI-based tools for a range of tasks, including simplifying medical and billing terminology for patients and allowing patients to self-schedule clinical visits.

“I think the earliest AI-enabled tools that are entering the rheumatology space would be AI-assisted computer vision adaptation for X-rays, a mainstream application pertinent to most rheumatology patients,” Dr. Mehta said. “For certain complex operations, like sharp scores for rheumatoid arthritis (RA), that are highly involved, time-consuming, low-yield tasks, even for seasoned radiologists, AI tools are a bit farther away.”

One of the challenges with using AI-enabled tools, explained Dr. Mehta, is the need for “very clean” data to generate useful validated predictive models. LLMs, such as ChatGPT, may be easier to implement for certain rheumatology clinical care domains.

“One application of AI is to provide physician-curated/-vetted responses to common patient questions, such as management of infection while on methotrexate therapy,” she said. “By streamlining tasks that would otherwise require communication with a rheumatology specialist or nursing staff, AI-assisted standard responses can reduce the workload on the clinical team and reduce burnout.”

While Dr. Mehta will highlight the benefits of using AI in rheumatology, she will also point out some of the challenges, such as its potential to exacerbate disparities if clinicians are not careful with how they integrate AI tools into practice.

Dr. Nelson will focus on four disparate use cases of AI-enabled tools in rheumatology imaging with potential near-term applicability in routine practice. One of the cases will review the role of AI in sacroiliac joint arthritis.

“It is exceedingly difficult to image that joint, and challenging and time-consuming to interpret the MRI scans,” Dr. Nelson said. “Therefore, there is an interest in exploring how AI can assist in image analysis on a large scale.”

Dr. Nelson will also discuss the role of AI in the analysis of pathology slides from patients with lupus nephritis and the pathology-based classification of patients into treatment categories.

“The technology is ahead of our ability to utilize it in a fair, equitable, and scientifically valid way,” Dr. Nelson said. “We all need to understand the assumptions and potential biases that are included in AI-based models to use them optimally.”

Along these lines, she will touch upon educational resources on AI handling of data, such as the National Institutes of Health (NIH) Big Data to Knowledge (BD2K) initiative and the Foundations of Biomedical Data Science YouTube recordings.

Ambient Listening

On Sunday, Nov. 17, from 10:30–11:30 a.m. ET in Room 146C, Michael Pfeffer, MD, Chief Information Officer and Associate Dean for Stanford Health Care and Stanford University School of Medicine, and Arinola Dada, MD, of the Overlake Arthritis and Osteoporosis Center,  will examine the uses and implementation of ambient listening technology to document natural conversations between providers and patients and reduce the after-visit workload of rheumatologists during AI: The Next Breakthrough in Documentation for Rheumatology.

The Tools and Challenges of AI

This year’s ARP Distinguished Lecture, AI in Rheumatology Practice – Unpacking the Toolkit, also will be delivered on Sunday, from 2:15–3:15 p.m. ET in Room 152B, by Jamie E. Collins, PhD, Biostatistician at the Orthopedic and Arthritis Center for Outcomes Research at Brigham and Women’s Hospital and Assistant Professor, Orthopedic Surgery, Harvard Medical School.

“AI has become ubiquitous, not only in clinical research and clinical care, but also in our day-to-day lives, with tools such as ChatGPT,” she said. “It is less a question of whether we want this technology in our lives, but a question of how do we make the most of it?”

Dr. Collins will talk broadly about how AI is being used in rheumatology care and research, provide a high-level overview of the potential benefits of using AI, and address drawbacks and concerns around AI in healthcare, particularly regarding the risk of bias and ethical issues related to data protection and privacy.

“AI has made huge strides in image analysis of data from X-rays, MRIs, and CT scans,” she said. “Google’s AI algorithm has been integrated into commercial mammogram analyses for breast cancer screening. In rheumatology, there have been exciting advancements in AI-assisted image processing, such as automated CT image processing to characterize cartilage or bone parameters in osteoarthritis.”

Personalized medicine in rheumatology is another potential area of AI innovation, Dr. Collins noted. Treatment individualization in management of rheumatologic conditions is a significant challenge.

“Currently, clinical judgment and non-clinical factors, such as insurance, inform the trial-and-error approach of therapeutic decision-making,” she said. “Researchers are now beginning to develop more personalized treatments, with improved response prediction, based on a breadth of data from various sources, including medical data warehouses and EHRs.”

While she envisions AI algorithms to assist in clinical decision-making, no AI algorithm is perfect, and algorithms are not expected  to make treatment plans independently, Dr. Collins emphasized.

A better understanding of the assumptions and variables that inform how AI algorithms generate outputs, and estimating the uncertainty in AI-assisted recommendations or predictions, can help improve a clinician’s comfort and uptake of these innovative tools in appropriate ways, she explained.

Using AI to Improve Medical Education

In the session Integrating AI into Rheumatology Education: A New Frontier for Trainees and Educators, on Tuesday, Nov. 19, from 9:30–10:30 a.m. ET in Room 102AB, Dinesh Rai, MD, Clinical AI Engineer, Boston Children’s Hospital, and Asit Misra, MD, Assistant Professor of Emergency Medicine at the University of Miami’s Gordon Center for Simulation and Innovation in Medical Education, will discuss how the latest AI innovations can expand medical education capabilities, and how AI-based diagnostic simulators can enhance clinical decision-making skills.

Dr. Rai, who has a mix of clinical and AI technical experience, will address the Two Sigma Problem — that students who receive individualized instruction perform two standard deviations higher than those taught in a conventional classroom.

“One of the most exciting parts of using AI in medical education in general is the potential for providing a ‘personal tutor’ to every student to address learner-specific gaps in their knowledge,” he said.

Dr. Rai will focus on modeling educational clinical cases, developed with clinical educators’ input, and then harnessing AI-assisted learning modules to tailor the case walk-throughs to an individual student’s learning profile.

The learning platform being developed by Dr. Rai and colleagues enables educators to incorporate other case-relevant learning resources, like journal articles and textbook chapters. The platform is agnostic to specialty or type of case; the educator can focus on different specialties, student education levels, and even learning goals.

The learning platform will also include a forum for students to explore the content without social stressors associated with traditional learning environments.