From the keynote lecture at the opening ceremony by Avi Goldfarb PhD, Rotman Chair in Artificial Intelligence and Healthcare at the University of Toronto entitled Prediction Machines: How AI Could Transform Health Care, to the last day of the meeting in a session titled, From Genotype to Phenotype: Towards Integration of Genetics and Clinical Medicine, with Drs. Katherine Liao and Yukinori Okada, there was a focus on the promise and occasional trepidation around the incorporation of predictive and generative artificial intelligence (AI) in rheumatic and musculoskeletal disease (RMD) research, education, and clinical practice at the recently concluded 2023 ACR Convergence.
Dr. Goldfarb’s keynote talk focused on generative AI and the future (a future perhaps more distant than portrayed in the popular media) of these technologies in health care writ large. More near-term, Dr. Liao discussed the work of her group from Brigham and Women’s Hospital in Boston which utilizes machine learning, LLMs, natural language processing, and other AI methodologies to develop algorithms allowing the use of EHR data for clinical research. The EHR holds vast promise for clinical research given the large numbers of patients and millions of available data points. However, there are many challenges to the use of these data, including how best to capture the patients of interest, how to manage the large resultant data set, how to overcome biases in the data (e.g., lack of access to care, fractured care, informative missingness), and how to bridge across siloed data from different systems and institutions to allow even larger analytic efforts with greater generalizability. For the latter, Dr. Liao introduced the concept of federated learning which can allow institutions to work with their data within their center, providing only aggregate results without identifiers to a central analytic hub, and thus reducing barriers presented both by concerns around patient privacy and representation as well as those around potentially limited expertise at any given center for achieving the final large scale analysis.
Dr. Okada, Professor of Statistical Genetics at Osaka University, discussed another application for advanced AI methodologies: in transomic analyses. These analyses allow incorporation of multiple large, disparate datasets on the same individuals (e.g., genetics, microbiome, metagenomics, metabolomics) to explore multi-faceted traits in human disease. He also discussed methods for cross-ancestry multi-trait genome-wide association studies, allowing a previously unavailable understanding of how genetic risk for disease might be similar or different among groups. (For context, those of white/European ancestry generally dramatically outnumber those of other groups, and if only one ancestry group can be studied at a time, analyses in non-White groups are often too underpowered to provide meaningful insights, if these individuals are studied at all.)
Precision medicine, an area that often utilizes AI methodologies like machine learning, has been an area of interest for rheumatologists, given our desire to target therapies to patients most likely to benefit and least likely to have adverse events in a less trial and error method than we currently use. Another outstanding talk by Professor Costantino Pitzalis (Queen Mary University of London), entitled Precision Clinical Trial Design in Rheumatoid Arthritis, provided an overview of the potential utility of precision medicine approaches for informing treatment decisions in rheumatoid arthritis (RA). Dr. Pitzalis and his group have utilized molecular signatures from ultrasound-guided synovial biopsies in RA clinical trials to better understand potentially important mechanistic subgroups, or endotypes. (As a side note, the use of ultrasound-guided synovial biopsy, as well as optimal use of ultrasound for other settings including inflammatory arthritis, soft tissue/periarticular issues, and procedural guidance, is the topic of several upcoming guidance documents by the ACR and USSONAR. Progress of these initiatives was discussed in two sessions at the meeting with a promise of results hopefully by the time of next years’ ACR Convergence in Washington D.C.) Dr. Pitzales and colleagues are now using these molecular signatures, with the help of new technology allowing rapid turnaround of the molecular results, to inform randomized clinical trials with the goal of improving response by giving the right treatment to the right patient at the right time.
A session that I was privileged to be a part of also focused on phenotyping and precision medicine, including in the EHR (Dr. Liao), osteoarthritis (myself), RMD pain (Dr. Tuhina Neogi from Boston University), and the resultant and very important implications of these subgroups on study design (particularly for clinical trials) by Dr. Jamie Collins from Harvard, through work funded by three NIAMS-funded Core Centers for Clinical Research (P30 CCCRs).
As Dr. Goldfarb noted, AI is here to stay, and although it will take time to discover its optimal role in our day to day work and the best way to incorporate it into education, its use in research is already moving rheumatology forward in remarkable and exciting ways.
Amanda E. Nelson, MD MSCR RhMSUS is an Associate Professor of Medicine at the University of North Carolina at Chapel Hill. Dr. Nelson has no conflicts of interest to report.
Image by GoodStudio / Shutterstock
