ARTIFICIAL INTELLIGENCE APPROACHES FOR RHEUMATOLOGY PRACTICE: RESULTS FROM AN INTERNATIONAL E-SURVEY OF THE DIGITAL RHEUMATOLOGY NETWORK

V. Venerito, L. Gupta, V. Agarwal, J. Knitza, T. D. R. N., T. HügleUniversity of Bari, Rheumatology Unit - Department of Precision and Regenerative Medicina and Ionian Area, Bari, Italy Royal Wolverhampton Hospitals NHS Trust, Department of Rheumatology, Wolverhampton, United Kingdom City Hospital, Sandwell and West Birmingham Hospitals NHS Trust, Department of Rheumatology, Birmingham, United Kingdom The University of Manchester, Division of Musculoskeletal and Dermatological Sciences, Centre for Musculoskeletal Research, School of Biological Sciences, Manchester, United Kingdom Sanjay Gandhi Postgraduate institute of medical sciences, Department of Clinical Immunology and Rheumatology, Lucknow, India Friedrich-Alexander University Erlangen-Nürnberg and Universitätsklinikum Erlangen, Department of Internal Medicine 3 - Rheumatology and Immunology, Erlangen, Germany The Digital Rheumatology Network, Working Group, Lausanne, Switzerland University Hospital Lausanne (CHUV), Switzerland, Department of Rheumatology, Lausanne, Switzerland  Background Artificial Intelligence (AI) is emerging as a promising tool to improve clinical decision-making and enhancing patient care. In rheumatology, AI may enable precision medicine, allowing patient profiling, prediction and treatment personalization [1]. Healthcare professionals’ (HPs’) perceptions regarding implementation of AI in rheumatology have been not investigated sufficiently so far. Objectives This international e-survey aimed to evaluate healthcare professionals’ attitudes towards AI’s potential benefits in managing patients with rheumatic diseases. Methods An online self-reported questionnaire (August 2022-December 2022) promoted using social media platforms was used to capture data on the knowledge of state-of-the-art AI methods. We evaluated attitudes, perceptions and expectations related to AI-powered clinical prediction modelling, computer vision and natural language processing (NLP) on e-health records (EHR). Univariate logistic regression was used to explore associations between demographic characteristcs of participants and their trust in machine learning and AI (Figure 1, panel A, expressed as a boolean.variable). Results Fifty-nine HPs (male n.36/58 respondents, 37.93%, 1 skipped) of mean age (±SD) 37.55 ± 10.12 years at a mean of 12.33 ± 10.54 years since graduation were considered for the analysis. Fifty-two out of 59 were specialists (88.14%), the largest part working in University Hospitals (45/58, 77.59%, 1 skipped). Thirty-seven out of 59 (62.71%) declared an open attitude in adopting machine learning tools for clinical prediction modelling for rheumatoid arthritis patients (Figure 1, Panel B). More than half of the respondents were interested in a prediction horizon of ≥1 year (33/59, 55.93%, Figure 1 Panel C). The vast majority (51/59, 86.44%) liked having an AI-powered clinical prediction embedded in EHRs, preferring disease activity scores as the target outcome (54/59, 91.53%). Consistently, most of the participants considered that clustering algorithms assigning patients to phenotypes have room in clinical practice, (45/59, 76.27%) especially in treatments selection (46/58, 79.31%, 1 skipped) and early arthritis management (37/59, 63.79%). Computer vision algorithms were considered of particular interest for the detection of erosions (39/59, 66.1%). Most of the participants agreed that machine learning could be beneficial to extract information from EHR (52/58, 89.66%). In particular, NLP was seen as a tool for capturing longitudinal changes among critical patient outcomes (36/54, 66.67%, 4 skipped) and for Reducing diagnostic delay by promptly identifying symptoms in primary care EHRs 36/58, 62.07%, 1 skipped). In general about an half of respondents (33/58, 56.90%), considered machine learning-based prediction as more powerful than conventional statistics in terms of clinical predictive modelling, especially for the potential of providing more sensitive analysis and identify smaller contributing variables (39/57, 68.42%). Neither age, gender, and time from degree were associated with trust in machine learning. Conclusion The participants of this international survey showed an open and optimistic attitude regarding AI implementation in clinical rheumatology. Expectations were mainly centered on improving patient management by allowing for accurate mid-to-long term prognosis. Reference [1]Venerito V, Angelini O, Fornaro M, Cacciapaglia F, Lopalco G, Iannone F. A Machine Learning Approach for Predicting Sustained Remission in Rheumatoid Arthritis Patients on Biologic Agents. J Clin Rheumatol. 2022 Mar 1;28(2):e334-e339. doi: 10.1097/RHU.0000000000001720. Image/graph:Figure 1. Participants answers to relevant questions. Acknowledgements We acknowledge the efforts of every member of the Digital Rheumatology Network. Disclosure of Interests None Declared. Keywords: Artificial intelligence DOI: 10.1136/annrheumdis-2023-eular.4429Citation: , volume 82, supplement 1, year 2023, page 1045Session: HPR Professional education, training and competencies (Poster View)