Artificial intelligence in sleep medicine I: Diagnosis, treatment, care, and research

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Not reported — no quantitative pooling performed; all findings are qualitative and descriptive

• Primary Outcome: Narrative synthesis of AI applications across sleep medicine including: (1) enhanced diagnostic accuracy in PSG and home sleep testing; (2) detection of early neurodegenerative changes in REM sleep behaviour disorder; (3) novel biomarker identification; (4) machine learning-based sleep apnea phenotyping for treatment selection; (5) CPAP adherence prediction; (6) NLP-facilitated clinical chart data extraction; and (7) chronotherapeutic timing optimisation

• Nnt Or Sensitivity: Not reported — no NNT, sensitivity, specificity, AUC, or hazard ratio data are synthesised at the review level; individual study metrics are discussed narratively without aggregation

• Confidence Interval: Not reported

Clinical Application

Feasibility is context-dependent and currently limited. AI-assisted PSG scoring tools are commercially available in some jurisdictions, but regulatory approval, local validation, and integration with existing clinical information systems remain significant barriers. CPAP adherence prediction models and NLP chart extraction tools are largely in research or pilot phases. The 'generalisation gap' identified by the authors — where models trained in one setting underperform in another — is a critical feasibility constraint for routine clinical deployment Australia has a well-developed sleep medicine infrastructure with RACGP and Thoracic Society of Australia and New Zealand (TSANZ) guidelines supporting home sleep testing and CPAP therapy for obstructive sleep apnoea. The TGA regulates AI-based medical devices under the Software as a Medical Device (SaMD) framework, and any AI diagnostic tool would require TGA conformity assessment before clinical deployment. PBS subsidises CPAP equipment for eligible patients with confirmed OSA (AHI ≥15 or AHI ≥5 with comorbidities), and AI-driven adherence prediction could theoretically support PBS compliance monitoring. One co-author (Leppanen) holds an affiliation with the University of Queensland, providing some Australian academic relevance. However, the review does not address Australian-specific datasets, Indigenous health equity considerations, or rural and remote access challenges — all of which are critical for equitable AI implementation in the Australian context. RACGP guidance on digital health and AI in general practice is evolving but does not yet specifically address sleep medicine AI tools. Adults with suspected or confirmed sleep disorders including obstructive sleep apnoea, REM sleep behaviour disorder, insomnia, and circadian rhythm disorders; applicable across primary care, respiratory medicine, neurology, and dedicated sleep medicine services

Abstract

Artificial intelligence (AI) is transforming sleep medicine (SM) through improved diagnostics, therapeutics, and research. AI enhances diagnostic accuracy, treatment personalization, workflow efficiency. However, implementation requires careful validation and oversight. This review explores AI applications across the sleep disorders spectrum. AI in diagnostic SM covers applications including polysomnography and home sleep testing. We examine how AI detects early neurodegenerative changes in REM sleep behavior disorder, identify novel biomarkers, and optimize chronotherapeutic timing. Wearables complement these advances by continuously monitoring sleep patterns, movement and physiological signals in natural settings, generating rich datasets ideal for AI analysis. In therapeutics, machine learning models enhance sleep apnea phenotyping, enabling precise treatment selection, and improve CPAP adherence prediction. AI helps identify biomarkers to optimize the timing of chronotherapeutic interventions. In medical care, natural language processing facilitates unstructured clinical chart data extraction. Implementation challenges include data standardization, algorithmic bias, and the "generalization gap." We provide a structured framework for clinical implementation, emphasizing validation requirements and ethical considerations. Large datasets using routine clinical care, cohort studies, and wearable derived data create vast amounts of data. In return, AI tools and methodologies enhance and expand development of phenotypes and endotypes with the final goal of personalized and precision SM.

References

1. 1.Sharafkhaneh, A., Hirshkowitz, M., Razjouyan, J., BaHammam, A., Leppanen, T., Shin, C., Korkalainen, H., & Penzel, T. (2026). Artificial intelligence in sleep medicine I: Diagnosis, treatment, care, and research. Sleep Medicine Reviews. Advance online publication. https://doi.org/10.1016/j.smrv.2026.102295