Applications and Challenges of Digital Health in Rheumatology

Rheumatic diseases encompass a broad spectrum of conditions, from common forms of arthritis to rare systemic autoimmune and autoinflammatory disorders such as systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA). Their diagnosis remains challenging due to clinical heterogeneity, overlapping features, and the rarity of some entities, often leading to delays in accurate identification and treatment initiation [1]. Digital health refers to the use of technologies such as artificial intelligence (AI), telemedicine, and electronic health records (EHRs) to improve health management efficiency, optimize treatment options, and enable more personalized care [2]. These technologies can facilitate early diagnosis and the development of personalized treatment plans, as well as optimize the allocation of resources to benefit more patients. Particularly in the current global context of uneven distribution of global health resources, the use of digital health tools can effectively compensate for a shortage of professionals and improve access to care. Although digital health technologies promise unprecedented efficiency gains and innovative breakthroughs, their potential limitations and ethical concerns cannot be ignored. The aim of this editorial is to explore the challenges of adopting digital health technologies in rheumatology practice and research, and to discuss the ethical issues raised by these innovations. Digital health, powered by AI, is transforming rheumatology through earlier diagnosis, personalized treatment, and proactive disease management, while complementary tools such as telemedicine, remote monitoring, and EHRs enhance care delivery, patient engagement, and overall disease management. Machine learning (ML) applied to genome-wide association study (GWAS) data has identified nine single nucleotide polymorphisms (SNPs) strongly associated with RA, enabling a high-performing polygenic risk score (PRS) model for screening high-risk individuals [3]. In imaging, deep learning (DL) models have achieved significant advances: a multi-stage DL algorithm can automatically predict the Sharp–van der Heijde (SvH) radiographic score from hand X-rays, improving both accuracy and efficiency compared to manual assessment [4]. Convolutional neural networks (CNNs) further extend this capability by detecting subtle abnormalities in X-ray, CT, and MRI scans relevant to inflammatory and structural joint damage (Table 1) [5]. Deep learning Confident learning Beyond AI, digital platforms are reshaping patient care. Telemedicine, via video consultations, mobile apps, and online portals, has expanded access for patients with RA and SLE, particularly during periods of restricted in-person care. The CareHand app, for instance, delivers tailored exercise programs and self-management support that significantly improve hand function, work performance, and patient satisfaction [6]. These platforms also enable remote monitoring of disease activity and treatment response, supporting more informed clinical decisions. Emerging large language models (LLMs) show potential to assist in symptom triage by allowing patients to input symptom duration, severity, and patterns to receive preliminary guidance, though their role remains supportive rather than diagnostic [7]. Figure 1 summarizes the current applications of digital health across the spectrum of rheumatology. Application of digital health in the field of rheumatology. Despite its promise, the integration of digital health into rheumatology faces multifaceted challenges spanning data quality, algorithmic robustness, clinical implementation, and ethical governance. Data limitations undermine model validity and equity. Table 2 summarizes the challenges and solutions for the application of digital health in rheumatology. GWAS data remain heavily skewed toward populations of European ancestry, drastically reducing the predictive performance of PRS in non-European groups [8]. Moreover, most machine learning models for rheumatic disease risk prediction rely solely on genomic data, neglecting critical gene-environment interactions and epigenetic influences. Addressing these gaps requires the inclusion of diverse cohorts and the integration of multi-omic and environmental data, ideally through cross-ethnic collaborative studies. Uneven and unrepresentative data sources Ignoring gene-environment interactions and epigenetic regulatory mechanisms Conducting cross-racial cohort studies Integrating environmental and epigenetic data MRI contrast differences by manufacturers Physician annotation inconsistencies Standardizing the preprocessing process Developing consensus for annotation Data privacy and security concerns Risk of poor decision-making Using federated learning or differential privacy Diagnostic basis visualization High shedding rate of wearable devices Lack of physiological parameter criteria Improving the comfort of wearable devices Adopting harmonized data standards Poor system interoperability Risk of data leakage Adopting the HL7 FHIR Establishing rheumatology AI alliance Technical heterogeneity compromises model generalizability. Even advanced CNNs struggle with variability in imaging protocols; for instance, differences in MRI magnetic field strength across manufacturers lead to inconsistent image contrast, limiting diagnostic reliability [9]. Similarly, wearable devices used in remote monitoring lack standardized metrics for physiological parameters (e.g., heart rate and activity levels), hindering cross-platform data integration [10]. Establishing consensus guidelines for data acquisition, annotation, and device interoperability is therefore essential to ensure reproducibility and clinical utility. Clinical integration remains incomplete. Although telemedicine has expanded access, particularly during public health emergencies [11], it cannot fully substitute for physical joint examinations or nuanced clinical assessments. Recognizing this, professional societies such as the European Alliance of Associations for Rheumatology (EULAR), the Asia Pacific League of Associations for Rheumatology (APLAR), and the American College of Rheumatology (ACR) have issued frameworks to guide its appropriate use. Likewise, patient adherence to wearables is often limited by factors such as discomfort or short battery life, underscoring the need to improve device usability alongside functionality. Ethical and infrastructural barriers persist. The deployment of LLMs raises concerns about data privacy, especially when handling sensitive health information. Techniques such as federated learning and differential privacy offer promising safeguards, but LLMs outputs must never replace clinical judgment. Instead, they should serve as decision-support tools that are transparent, explainable, and validated in real-world settings [12]. Concurrently, poor interoperability between EHRs systems impedes seamless data exchange across institutions. The adoption of standards such as HL7 Fast Healthcare Interoperability Resources (FHIR) can significantly enhance compatibility [13], yet cloud-based data storage continues to pose cybersecurity risks [14]. To accelerate progress, we propose the formation of an international rheumatology AI alliance dedicated to developing harmonized data-sharing protocols, benchmarking frameworks, and ethical guidelines. The rapid integration of digital health into rheumatology raises profound ethical considerations that span individual rights, clinical accountability, and global governance. At the individual level, robust safeguards are essential to protect patient autonomy and privacy. Before any research or clinical deployment, formal approval from an ethics review board and obtaining truly informed consent must be secured [15]. Given the sensitivity of health data, particularly granular information on symptoms, genomics, or disease trajectories, stringent technical and organizational measures (e.g., encryption, access controls) are required to prevent unauthorized access or misuse. Moreover, patients should be clearly informed about how their data will be collected, used, stored, and shared through transparent and accessible data governance policies [16]. Crucially, when AI-informed recommendations influence treatment decisions, patients must retain the right to understand, question, and ultimately consent to or decline such options, thereby preserving shared decision-making as a cornerstone of ethical care. Regarding accountability, the delegation of clinical support to AI systems complicates traditional models of liability. Although AI does not act autonomously, errors arising from flawed algorithms, biased training data, or inappropriate clinical implementation can lead to patient harm. A clear framework for attributing responsibility, which clarifies the roles of developers, healthcare institutions, clinicians, and regulators, is therefore essential to resolve disputes, ensure redress, and uphold patient rights. On the global scale, cross-border data sharing, which is critical for training generalizable models, faces significant legal and technical fragmentation. Divergent regulatory regimes (e.g., GDPR in Europe and HIPAA in the U.S.) and incompatible data standards heighten risks related to privacy, sovereignty, and equity [17]. Rather than seeking full harmonization, the international community should prioritize interoperable data protection frameworks and mutual recognition agreements that enable secure, ethical data flows while respecting jurisdictional differences. Initiatives such as federated learning and trusted research environments offer promising pathways to advance global collaboration without compromising local compliance. Digital health presents transformative potential for rheumatic disease care—enabling earlier diagnosis, personalized risk prediction, and broader access through telemedicine. Realizing this potential requires overcoming key technical and ethical challenges through interdisciplinary collaboration and global governance to ensure equitable and effective care for all. Xiao-Xiao Li: conceptualization, formal analysis, investigation, writing – original draft. Xue-Er Cheng: data curation, methodology, software, writing – review and editing. Ming-Ming Zhang: supervision, visualization. Peng Wang: project administration, writing – review and editing. Hai-Feng Pan: resources, validation, funding acquisition. All authors have read and approved the final manuscript. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. This study was funded by grants from the National Natural Science Foundation of China (Grants 82273710 and 82404354), Research Funds of Center for Big Data and Population Health of IHM (Grant JKS2022017), and Research Innovation Team Project of School of Public Health, Anhui Medical University (Grant kctd200402). The authors declare no conflicts of interest. The authors have nothing to report.