2025 AASM立场声明：睡眠医学中的人工智能（更新版）

SPECIAL ARTICLES

# Artificial intelligence in sleep medicine: an updated American Academy of Sleep Medicine position statement

Margarita Oks, MD$^{1}$; Ramesh Sachdeva, MD, PhD$^{2}$; Mattina A. Davenport, PhD$^{3}$; Aatif M. Husain, MD$^{4}$; Maninder Kalra, MD$^{3}$; Vidya Krishnan, MD, MHS, FAASM$^{5}$; Trung Le, PhD$^{6}$; Ankit A. Parekh, PhD$^{7}$; Hyung Park, MD$^{8}$; Bharati Prasad, MD$^{9}$; Scott M. Ryals, MD, FAASM$^{10}$; Subaila Zia, MD, MBA, FCCP$^{11}$; Matthew Anastasi, RPSGT$^{12}$; Anita V. Shelgikar, MD, MHPE$^{13}$; Fariha Abbasi-Feinberg, MD$^{14}$; Alexandre Rocha Abreu, MD$^{15}$; Anuja Bandyopadhyay, MD$^{16}$; Indira Gurubhagavatula, MD, MPH$^{17}$,$^{18}$; Vishesh K. Kapur, MD, MPH$^{19}$; David Kuhlmann, MD$^{20}$; Eric J. Olson, MD$^{21}$; Susheel P. Patil, MD, PhD$^{22}$,$^{23}$; James A. Rowley, MD$^{24}$; Lynn Marie Trotti, MD, MSc$^{25}$; Emerson M. Wickwire, PhD$^{26}$,$^{27}$

$^{1}$Northwell Health System, New York, New York; $^{2}$Children's Hospital of Michigan, Detroit, Michigan; $^{3}$Nationwide Children's Hospital, Columbus, Ohio; $^{4}$Duke University Medical Center, Durham, North Carolina; $^{5}$MetroHealth Medical Center, Cleveland, Ohio; $^{6}$University of South Florida, Tampa, Florida; $^{7}$Icahn School of Medicine at Mount Sinai, New York, New York; $^{8}$Providence Medical Group INWA, Spokane, Washington; $^{9}$University of Illinois Chicago, Chicago, Illinois; $^{10}$Atrium Health, Matthews, North Carolina; $^{11}$Telemedora, Sunnyvale, California; $^{12}$American Academy of Sleep Medicine, Darien, Illinois; $^{13}$University of Michigan, Ann Arbor, Michigan; $^{14}$Sleep Medicine, Millennium Physician Group, Fort Myers, Florida; $^{15}$Division of Pulmonary, Critical Care and Sleep Medicine, Miller School of Medicine, University of Miami, Miami, Florida; $^{16}$Department of Pediatrics, Indiana University School of Medicine, Indianapolis, Indiana; $^{17}$Division of Sleep Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; $^{18}$Corporal Michael J. Cressee Veterans Affairs Medical Center, Philadelphia, Pennsylvania; $^{19}$Division of Pulmonary, Critical Care and Sleep Medicine, University of Washington, Seattle, Washington; $^{20}$Sleep Medicine, Bothwell Regional Health Center, Sedalia, Missouri; $^{21}$Division of Pulmonary and Critical Care Medicine, Center for Sleep Medicine, Mayo Clinic, Rochester, Minnesota; $^{22}$Sleep Medicine Program, University Hospitals of Cleveland, Cleveland, Ohio; $^{23}$Case Western Reserve University School of Medicine, Cleveland, Ohio; $^{24}$Rush University Medical Center, Chicago, Illinois; $^{25}$Emory Sleep Center and Department of Neurology, Emory University School of Medicine, Atlanta, Georgia; $^{26}$Division of Pulmonary, Critical Care, and Sleep Medicine, Department of Medicine, University of Maryland School of Medicine, Baltimore, Maryland; $^{27}$Department of Psychiatry, University of Maryland School of Medicine, Baltimore, Maryland

The field of artificial intelligence (AI) is rapidly expanding and has the potential to significantly impact the practice of sleep medicine. Sleep practitioners may be able to augment the efficiency and effectiveness of care delivery to patients by incorporating AI into clinical practice. However, AI must be implemented in a responsible manner, with a full appreciation of the technology's current limitations and evolving nature. Specifically, the tenets of data privacy, fairness and transparency, infrastructure, and medical-legal issues must be considered. Many of these issues are evolving and will continue to impact AI implementation in new ways. Enhancements in technology will provide new AI options for clinicians and patients. Evolving case law will influence the legal risks that clinicians, hospitals, and sleep centers will face. Creation of new policies within the United States and internationally will provide frameworks and safeguards for data privacy and AI use. Sleep clinicians and researchers should remain updated and knowledgeable in these areas. As AI continues to develop and advance in the sleep medicine field, it must be a partner to, and not a replacement for, clinician (ie, human) oversight. It is the position of the American Academy of Sleep Medicine that responsible AI integration into sleep medicine has the potential to enhance clinical care and research but requires careful consideration to overcome clinical validation challenges, ensure ongoing accuracy after implementation, and incorporate clinically relevant and user-friendly tools into practice, while also upholding standards of safety, appropriateness, and transparency.

Keywords: artificial intelligence, clinical care, research, patient safety

Citation: Oks M, Sachdeva R, Davenport MA, et al. Artificial intelligence in sleep medicine: an updated American Academy of Sleep Medicine position statement. J Clin Sleep Med. 2025;21(11):1953–1955.

INTRODUCTION

Since 2020, when the first American Academy of Sleep Medicine position statement on artificial intelligence (AI) was released,$^{1}$ the potential use of AI in sleep medicine has evolved at lightning speed with significant technological advancements. AI is defined as the ability of machines to learn through experience and gain proficiency through examples, not just by carrying out computations as instructed by human users.$^{2,3}$ AI implementation has grown beyond attended polysomnography scoring and has the potential to aid clinical documentation, predict risk of having a disease, diagnose conditions, choose optimal treatments, and predict and monitor treatment response and clinical outcomes. Although AI technology has continued to expand, the appropriate utilization

of AI in clinical sleep medicine has been slower to catch up due to various factors related to implementation. Privacy, fairness and transparency, infrastructure, and medical-legal components should be individually addressed to ensure that AI is used appropriately, is applied safely, and truly begins to serve as an integrated adjunct to sleep medicine clinical care and research with continued human oversight and input.

POSITION

It is the position of the American Academy of Sleep Medicine that responsible AI integration into sleep medicine has the potential to enhance clinical care and research but requires careful

M Oks, R Sachdeva, MA Davenport, et al.

Artificial intelligence in sleep medicine

consideration to overcome clinical validation challenges, ensure ongoing accuracy after implementation, and incorporate clinically relevant and user-friendly tools into practice, while also upholding standards of safety, appropriateness, and transparency.

DISCUSSION

There are 4 main categories for consideration when integrating AI into sleep medicine: data privacy, fairness and transparency, infrastructure, and medical-legal tenets.

Data privacy

Data security reinforces data privacy by preventing unauthorized use of patient information. Security challenges arise from the storage and processing of large volumes of medical data. Data security is essential, as patient data can potentially be used for AI training models without patients' knowledge.$^{4}$ Sleep medicine clinical and research entities that incorporate AI should conduct risk assessments, report data leakages, and follow other best practices for data security. Sleep medicine entities must adhere to institutional and governmental regulations and implement related workflows and policies to ensure data are protected.$^{5}$

# Fairness and transparency

Development and implementation of equitable AI-powered sleep technologies warrants a multidimensional approach from different stakeholders who remain engaged and adaptive to the rapidly evolving stratosphere of digital health and needs of society.$^{6,7}$ Datasets used to train AI/machine learning algorithms should be evaluated for bias before AI-based sleep technologies are deployed in practice.$^{7,8}$ A consistent approach should be used to evaluate and assess bias in all AI-based sleep technologies.$^{2,8-10}$ Field standards for evaluation of data representatives (race/ethnicity, sex, socioeconomic status, disability status, etc.) and bias in available sleep datasets (eg, electronic health records, wearable technology output public datasets) should be created.$^{4,7,8,11-13}$ Continued validation of AI training models to ensure all patient groups are included in assessments is essential.$^{4,9,11-14}$ Access to remote home monitoring/user-generated data for health care integration should be prioritized while remaining cognizant of barriers that limit access to health care. Monitoring systems should be implemented by health care and resource-sharing entities to ensure that AI systems remain relevant, effective, and equitable after commercial deployment.$^{4,9,15,16}$ Due to the widening digital divide, whereby areas with limited access to technology may not be represented in datasets or gain access to care, ethical considerations at all levels of AI deployment are integral to the development of a fully trusted, transparent, and successful system that minimizes health disparities.$^{4,7,8,10-13,15,16}$

Infrastructure

As AI continues to expand and become mainstream in sleep medicine, infrastructure enhancements should be considered an integral part of any AI implementation. Use of AI will lead to the development of new and updated testing methods as well as technical standards; sleep facilities will need to continually assess how these changes may impact their centers. Organizations will

need to be equipped with both on-site and cloud-based computing systems and personnel who are well versed in AI terminology so that new standards and capabilities can be utilized and effectively propagated to all stakeholders. Due to an expected increase in data generation with the use of AI, improved data storage systems that are Health Insurance Portability and Accountability Act compliant and have cyber intrusion monitoring in place should be deployed. Collaboration among sleep researchers, developers, and commercial entities to standardize AI software will allow for quicker and more efficient implementation of AI across all interested organizations.

Medical-legal tenets

Responsible integration of AI in sleep medicine requires thoughtful consideration of potential medical-legal implications. Food and Drug Administration approval provides a regulatory process that aims to ensure the safety and efficacy of a specific device and its components.$^{14}$ Knowledge of applicable laws and regulations (or lack thereof) is now paramount, and the legislative and regulatory void will need to be filled as AI advances. Use of AI for medical diagnosis and clinical decisions in the absence of human supervision prompts the question of liability and accountability for patient harm, which has implication for medical malpractice, and hence should never be used without human oversight.$^{2,16-18}$ The importance of algorithmic transparency and explainable AI is crucial; software that learns and changes over time may make it difficult to find accountability if harm arises.$^{2,17,18}$ In 2024, the American Medical Association released their guidance on AI policies in health care. Specifically, this policy highlighted the following themes: (1) need for governance of AI with an emphasis on transparency and disclosure (this is of even greater importance while using generative AI to minimize risks); (2) recognition of issues related to physician liability, data privacy, and cybersecurity; and (3) role of medical specialty societies to “set the standards” for AI use in their specific domain.$^{19,20}$ Federal, state, and local legislatures should start to focus on developing laws regarding patient-related disclosure if and when AI is used for clinical care, including data interpretation, patient messaging, and clinical advice.$^{19-21}$

FUTURE DIRECTIONS

The future integration of AI in sleep medicine is expected to advance in several key areas: (1) personalized screening, diagnosis, and treatment; (2) embedding AI technologies in health care infrastructure; (3) adopting AI models for clinical decision-making; and (4) innovating clinical management. The goals could focus on improving sleep disorder diagnostics and treatment by incorporating genomic data, molecular tests, and imaging data into AI models to enhance proactive care. These diagnostic and treatment goals need to be developed with attention to social determinants of health, data equity, and access. AI tools may be able to predict the onset and progression of comorbidities like metabolic, neurologic, and cardiovascular diseases based on sleep data.$^{22}$ AI-based technologies like edge computing, wearables, telemedicine, and robotics may enhance health care access, providing continuous symptom tracking and follow-up care.$^{23,24}$ The use of wearable technologies for sleep monitoring offers the opportunity to

M Oks, R Sachdeva, MA Davenport, et al.

Artificial intelligence in sleep medicine

identify longitudinal changes in physiology.$^{25}$ Generative and reasoning AI models, powered by domain-specific knowledge, will analyze neuroimaging, genetic, and biomolecular data, and create synthetic data from existing datasets, offering extended insights into sleep health.$^{22,26}$ Interactive AI decision support systems may help clinicians with real-time diagnosis and treatment recommendations.$^{27}$ AI may also streamline clinical workflows by linking medical records and biological data, automating tasks such as documentation and between-visit care. This integration may allow sleep medicine clinicians to deliver more precise patient care, with the potential to improve both provider satisfaction and patient outcomes. Every step of continued AI evolution in sleep medicine should be supported by clinical, ethics, legal, and research experts whose goal should be to ensure appropriate AI development and implementation.$^{7}$ AI should be an adjunct to, and not a replacement for, clinician involvement in all aspects of sleep medicine patient care and research development.

CONCLUSIONS

The use of AI in sleep medicine should aim to move beyond simple diagnostic assistance and become a part of clinical practice and research to elevate and improve the care provided to patients. To ensure appropriate integration of AI in sleep medicine, privacy, fairness and transparency, infrastructure, and medical-legal tenets should be addressed, developed, and adapted.

REFERENCES

1. Goldstein CA, Berry RB, Kent DT, et al. Artificial intelligence in sleep medicine: an American Academy of Sleep Medicine position statement. J Clin Sleep Med. 2020; 16(4):605-607. 2. Bandyopadhyay A, Goldstein C. Clinical applications of artificial intelligence in sleep medicine: a sleep clinician's perspective. Sleep Breath. 2023;27(1):39-55. 3. National Institute of Biomedical Imaging and Bioengineering. Artificial intelligence (AI). https://www.nibib.nih.gov/science-education/science-topics/artificial-intelligence ai. Published March 2025. Accessed July 1, 2025. 4. Marques M, Almeida A, Pereira H. The medicine revolution through artificial intelligence: ethical challenges of machine learning algorithms in decision-making. Cureus. 2024;16(9):69405. 5. Hiter S. AI and privacy issues: challenges, solutions, and best practices. eWeek. June 20, 2024. 6. Watson NF, Fernandez CR. Artificial intelligence and sleep: advancing sleep medicine. Sleep Med Rev. 2021;59:101512. 7. Verma RK, Dhillon G, Grewal H, et al. Artificial intelligence in sleep medicine: present and future. World J Clin Cases. 2023;11(34):8106-8110. 8. Hallowell N, Parker M, Nellåker C. Big data phenotyping in rare diseases: some ethical issues. Genet Med. 2019;21(2):272-274. 9. Embi P. Algorithmovigilance-advancing methods to analyze and monitor artificial intelligence-driven health care for effectiveness and equity. JAMA Netw Open. 2021;4(4):e214622.

10. Zhang J, Whebell S, Gallifant J, et al. An interactive dashboard to track themes, development maturity, and global equity in clinical artificial intelligence research. Lancet Digit Health. 2022;4(4):e212-813. 11. Webb EK, Etter JA, Kwasa JA. Addressing racial and phenotypic bias in human neuroscience methods. Nat Neurosci. 2022;25(4):410-414. 12. Kawai M, O'Hara R, Einen M, Lin L, Mignot E. Narcolepsy in African Americans. Sleep. 2015;38(11):1673-1681.

13. Spruyt K. Narcolepsy presentation in diverse populations: an update. Curr Sleep Med Rep. 202;6(4):239–250. 14. Bandyopadhyay A, Bae C, Cheng H, et al. Smart sleep: what to consider when adopting AI-enabled solutions in clinical practice of sleep medicine. J Clin Sleep Med. 2023;19(10):1823–1833. 15. Thomasian NM, Eickhoff C, Adashi EY, et al. Advancing health equity with artificial intelligence. J Public Health Policy. 2021;42(4):602–611. 16. Naik N, Zeeshan Hamedd BM, Shetty DK, et al. Legal and ethical consideration in artificial intelligence in healthcare: who takes responsibility? Front Surg. 2022;9:862322. 17. Bandyopadhyay A, Oks M, Sun H, et al. Strengths, weaknesses, opportunities, and threats of using AI-enabled technology in sleep medicine: a commentary. J Clin Sleep Med. 2024;20(7):1183–1191. 18. Čartolovni A, Tomići A, Lazić Mosler E. Ethical, legal, and social considerations of AI-based medical decision-support tools: a scoping review. Int J Med Inform. 2022; 161:104738. 19. Medical Board of California. Generative artificial intelligence (GenAI) notification requirements. https://www.mbc.ca.gov/Resources/Medical-Resources/GenAI-Notification.aspx. Published 2025. Accessed June 3, 2025. 20. American Medical Association. Augmented intelligence development, deployment, and use in health care. https://www.ama-assn.org/system/files/ama-ai-principles.pdf. Published November 2024. Accessed June 3, 2025. 21. Cartolovni A, Malešević A, Poslon L. Critical analysis of the AI impact on the patient-physician relationship: a multi-stakeholder qualitative study. Digit Health. 2023;9:20552076231220833. 22. Baizer L, Bures R, Nadkarni G, et al. Big data approaches for novel mechanistic insights on sleep and circadian rhythms: a workshop summary. Sleep. 2025;48(6): sazf035. 23. De Zambotti M, Goldstein C, Cook J, et al. State of the science and recommendations for using wearable technology in sleep and circadian research. Sleep. 2024;47(4):zsad325. 24. Le TQ, Huynh P, Tomeselli L. Navigating the night: evaluating and accessing wearable sleep trackers for clinical use. Sleep. 2024;47(3):zsad319. 25. Yoon H, Choi SH. Technologies for sleep monitoring at home: wearables and nearables. Biomed Eng Lett. 2023;13(3):313–327. 26. Zhu L, Mou W, Xie J, Luo P, Chen R. What is the best approach to assessing generative AI in medicine? Resuscitation. 2024;197:110164. 27. Platz JJ. Generative AI in clinical medicine: opportunity for improvement. Ann Thorac Surg. 2024;118(6):1342.

ACKNOWLEDGMENTS

The American Academy of Sleep Medicine Artificial Intelligence Committee and Board of Directors thank Andrew Sampson, MBA, who assisted with the development of this statement.

SUBMISSION & CORRESPONDENCE INFORMATION

Submitted July 28, 2025 Accepted July 28, 2025 Address correspondence to: Margarita Oks, MD, Northwell Health System, 100 East 77th Street, New York, NY 10075; Email: moks@northwell.edu

Address correspondence to: Margarita Oks, MD, Northwell Health System, 100 East 77th Street, New York, NY 10075; Email: moks@northwell.edu

DISCLOSURE STATEMENT

The authors are the 2024–2025 members of the American Academy of Sleep Medicine (AASM) Artificial Intelligence in Sleep Medicine Committee and the 2024–2025 members of the AASM Board of Directors. This statement is published by the AASM as an advisory that is to be used for educational and informational purposes only. A.M.H. is a consultant for Pranu, Merk, Marinus Pharmaceuticals, LivaNova USA Inc., UCB Inc., UniQURE, Jazz Pharmaceuticals, and EISAI Inc. A.A.P. is a consultant for Cerebra Medical, Greenkey Health, and Rajant Health. R.S. is a consultant for Planatome. M.A. is employed by the American Academy of Sleep Medicine.