Patient experiences, attitudes, and profiles regarding artificial intelligence in rheumatology: a German national cross-sectional survey study

Abstract

While artificial intelligence (AI) is gaining attention in rheumatology, little is known about patient perspectives. This study addresses this gap by examining patients' experiences and attitudes toward AI. A nationwide, cross-sectional, web-based survey was conducted between March and May 2025 among adult patients with rheumatic diseases in Germany. Data were analyzed descriptively and with cluster analysis. A total of 778 patients completed the survey (70.4% female, mean age 51.3 years). The most common diagnosis was rheumatoid arthritis (31.7%). While 26.8% reported current AI use for health-related purposes, 57.8% expressed interest in using it. Patients were particularly interested in AI-based symptom checkers (64.3%), therapy recommendations (50.6%), and chatbots for medical inquiries (44.5%). 57.6% of patients indicated that they would welcome their rheumatologists using AI-based clinical suppport. The most frequently cited benefits of AI included improved information access (63.5%) and faster diagnosis (57.7%), while concerns centered on faulty AI (74.3%) and reduced human interaction (59.6%). Cluster analysis identified three distinct patient profiles: 'AI-savvy' (41.4%), 'AI-pragmatic' (44.8%), and 'AI-skeptical' (13.8%). Cluster membership was significantly associated with age and education, with younger patients more often belonging to the 'AI-savvy' group. Patients with rheumatic diseases showed substantial interest in AI-supported care, although actual use in medical contexts remained limited. Age and education differences highlight the need for tailored implementation strategies to ensure equitable and patient-centered adoption of AI in rheumatology.

Keywords: Artifical intelligence; ChatGPT; Large language models; Patient self-management; Surveys and questionnaires.

Fig. 1

Sankey diagram to illustrate A usage of AI-based tools for health-related use cases, B interest in AI and willingness to C donate data for AI research, D use AI-based self-management and E employ AI-based disease monitoring using AI and wearables. (N = 777–778). The discrepancy in the sample size is due to one missing answer. Green pathways show positive responses, red shows negative, and grey shows neutral or undecided responses. The Sankey diagram was created using the SankeyMATIC online tool. Pro and con in Fig. 1E represent favorable and unfavorable attitudes toward AI-based disease monitoring

Fig. 2

Current and Potential use of AI in healthcare and patient attitudes toward AI-assisted care. A shows in green how often patients currently use AI for specific healthcare purposes (N = 778). B shows the average perceived usefulness (mean ± SD) of AI among patients (N = 740). C shows the percentage of patients in light green who think that specifically defined rheumatology AI tools could be useful (N = 778). AI-supported automatic joint ultrasound’, ‘AI-supported automatic blood sampling’, and ‘AI-supported chatbot’ refer to prototype systems demonstrating artificial intelligence applications in musculoskeletal diagnostics, automated phlebotomy, and patient communication, respectively. D shows patients’ opinion on the use of AI-supported second opinion by physicians (N = 778).The figure was created using R and RStudio

Fig. 3

Perceived advantages, barriers and facilitators of AI in rheumatology. A shows the proportion of patients who recognize specific advantages of AI. B illustrates key concerns hindering AI adoption. C presents factors that could encourage AI use. Each stacked bar reflects the percentage of respondents who selected each item (N = 778). The figure was created using R and RStudio. The item “improved communication with doctors” was not further specified (e.g. frequency, quality, or efficiency) in the survey

Fig. 4

A Patient clusters and B their respective characteristics. By clustering analysis, three distinct patient groups based on their use of and attitudes toward AI in healthcare were identified. A shows clustering via Principal Component Analysis (PCA) scatterplot which illustrates the spatial distribution of patients across two principal dimensions derived from the data. Three distinct clusters are identified and visually separated using colored ellipses: cluster 1 – AI-savvy (n = 306): green, cluster 2 – AI-skeptical (n = 102): orange, cluster 3 – AI-pragmatic (n = 331): blue. B The heatmap shows differences between the three clusters across multiple variables by mean scores. Significant differences are indicated by p-values and confidence intervals (CI), N = 739. The graphs were created using R and RStudio

Fig. 5

Age and education level distribution across clusters. A shows the figure legend. B displays the percentage of individuals in each cluster—AI-savvy (green), AI-skeptical (orange), and AI-pragmatic (blue)—across the three age groups: 18–39, 40–59, and 60 + years. C shows how education levels are distributed within each cluster, ranging from no certificate to university degree. Statistically significant differences (p < 0.05) between clusters are indicated, N = 739. The graphs were created using R and RStudio