Use and Perceived Accuracy of Artificial Intelligence (ChatGPT) for Self-Leaning in Inflammatory Arthritis: Opportunities for Equitable Access to Appropriate Information?

Le résumé fourni par la source

Objectives Our study identified demographic, clinical, and educational factors associated with using ChatGPT (an artificial intelligence, AI application) for self-learning, and the perceived accuracy of information provided by ChatGPT among adults with inflammatory arthritis (IA). Methods Between March and July 2025, we conducted a cross-sectional survey of 410 adults with IA followed at the McGill University Health Centre. Self-reported diagnoses included rheumatoid arthritis (61.5%), spondyloarthritis (13.4%), psoriatic arthritis (9%), systemic lupus erythematosus (7.3%), juvenile idiopathic arthritis (5.1%), and others (3.2%). The primary outcome ‘frequency of AI/ChatGPT use to inquire about their medical condition’ was categorized as an ordered outcome (never rarely/occasionally; somewhat /very frequently) and alternatively as a dichotomous outcome (never used vs ever used AI). The secondary outcome ‘accuracy of information provided by AI’ was categorized dichotomously as rarely/never accurate vs always/mostly accurate. Potential correlates included sociodemographic characteristics, disease duration, prior access to educational resources, reported barriers to education, prior educational resources, and preference for online formats. We built multivariate logistic regressions with either ordered or dichotomous outcomes. Results Participants were predominantly middle-aged (mean ± SD 49.6 ± 18.8 years), female (72.7%), urban residents (87.3%), with established IA (77.6%, >5 years). Overall, 59% reported never using AI, 31.7% used it rarely/occasionally, 8.5% used it somewhat/very frequently. In multivariate ordered (logit) models, older age (per-year aOR 0.64, 95% CI 0.51-0.79), established IA (>5 years; aOR 0.51, 95% CI 0.32-0.83), and prior arthritis education (aOR 0.58, 95% CI 0.33-0.99) were associated with lower AI use, while preference for online formats for educational resources predicted greater AI use (aOR 2.06, 95% CI 1.28-3.36). In dichotomous logistic models, female sex (aOR 0.61, 95% CI 0.38-0.97), older age (per-year aOR 0.98, 95% CI 0.97-0.99), and established IA (>5 years; aOR 0.54, 95% CI 0.32-0.89) were associated with lower odds of AI use, whereas online preference again correlated with AI use (aOR 2.25, 95% CI 1.39-3.71). Among AI users, older age was associated with lower perceived information accuracy (per-year aOR 0.96, 95% CI 0.94-0.98). No other covariates were significant. Conclusion Many patients with IA use ChatGPT for self-learning, especially younger ones and those preferring online resources. In contrast, women, older patients, those with long-standing disease, or relying on traditional education were less likely to use it. Among users, older age was associated with lower perceived accuracy. Future work should explore how to best integrate AI while ensuring accuracy and equity.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.