JMIR Publications

ABSTRACT

Artificial intelligence (AI) is transforming diabetes mellitus (DM) care by enabling earlier detection, precision management, and data-driven decision support across clinical settings. However, the growing number of systematic reviews has produced a fragmented evidence base, making it difficult to assess the field’s maturity, methodological quality, and clinical readiness, and leaving future research directions largely undefined. To clarify this landscape, we conducted a tertiary review that synthesizes 70 systematic reviews on AI applications across type 1, type 2, and gestational diabetes and their major complications. Our analysis reveals an accelerating research trajectory since 2020, yet activity remains concentrated in type 1 and type 2 diabetes and in image-based applications such as diabetic retinopathy and foot-ulcer screening, with gestational and kidney disease still underexplored. Methodological rigor is inconsistent, with wide variation in search coverage, quality appraisal, and reporting standards. Across the literature, recurrent challenges include data bias, limited external validation, and barriers to clinical implementation. By integrating evidence across disease types, data modalities, and levels of care, this review shows that AI in diabetes is at a transitional stage. The field is technically advanced yet remains clinically fragmented and lacks a coherent roadmap for translation into practice. Expanding research into underrepresented areas such as multimodal data integration, natural-language processing of clinical text, and reinforcement learning for adaptive decision support will be essential to bridge this gap. This synthesis consolidates over a decade of secondary evidence and clarifies the structural and methodological requirements for achieving trustworthy, equitable, and scalable AI-driven diabetes care.