JMIR Publications

ABSTRACT

Background:

Sleep disorders represent a significant public health burden associated with cardiovascular and neurocognitive morbidities. While artificial intelligence (AI) technologies, ranging from deep learning diagnostic algorithms to large language models (LLMs), offer potential for personalized sleep medicine, their integration into clinical practice remains limited. This translational disparity is often attributed to a lack of human-centered design, specifically insufficient stakeholder engagement in the development and implementation of these technologies. Current research frequently prioritizes algorithmic performance metrics over usability, clinical workflow integration, and patient trust.

Objective:

This scoping review aims to systematically characterize the extent and nature of Human-Centered AI (HCAI) research within sleep medicine. The study seeks to map the distribution of research activities across different AI modalities and stakeholder engagement strategies, identify methodological disparities, and evaluate how diverse stakeholders are involved in the design, validation, and implementation of AI tools including patients, clinicians, and technologists.

Methods:

We conducted a scoping review following the PRISMA-ScR guidelines. A comprehensive search was performed to identify studies describing the design, development, or evaluation of AI technologies for sleep health with explicit human-centered components. Included studies (n=32) were categorized based on AI technology type and the method of stakeholder engagement.

Results:

Based on the 32 included studies, the analysis reveals a significant maldistribution of research focus across technological domains. Research on Generative AI is predominantly restricted to downstream expert auditing of output accuracy (comprising 78% of Generative AI studies), with a marked deficiency in upstream participatory design involving patients. Conversely, Deep Learning research is heavily concentrated on technical Explainable AI (XAI) methods to address algorithmic opacity for clinicians, yet lacks progression to real-world clinical implementation. mHealth and Wearable technologies (n=17, 53%) demonstrate the most balanced HCAI ecosystem, evidencing a complete translational cycle from upstream Co-design to downstream Clinical Implementation. Furthermore, a paradigm shift is observed wherein AI is evolving from an automated diagnostic tool into an interactive therapeutic agent, with recent studies indicating that lay users may perceive responses from LLMs as more empathetic than those from physicians.

Conclusions:

Current HCAI research in sleep medicine is characterized by distinct methodological disparities and ecological validity gaps. To advance the field, future research must prioritize a participatory approach that engages stakeholders early in the algorithm development process, particularly for Generative AI. Future studies must also address current demographic biases through inclusive recruitment and validate long-term adherence in diverse clinical populations. Additionally, there is a critical need to apply implementation science frameworks to translate high-performance deep learning models into clinical practice. Establishing standardized protocols for human-AI interaction and ensuring algorithmic transparency are essential for fostering the clinical trust and safety required for the effective adoption of AI in sleep health management.