EUVIMEDEUVIMED entdecken
Crossref

Are multimodal artificial intelligence predicting proxies for age and sex in sleep-based cardiovascular risk models?

Benjamin Fox iD, Girish N Nadkarni, Ankit Parekh iD

SLEEP · 08.10.2026

DOI: 10.1093/sleep/zsag266 ↗

Kurz gefasst

Fachartikel in der Zeitschrift SLEEP, online erschienen am 08.10.2026. Verfasst von Benjamin Fox und 2 weiteren Autor:innen. Beteiligte Einrichtung: The Windreich Department of Artificial Intelligence and Human Health, Icahn School of Medicine at Mount Sinai, New York, NY, USA. Die Arbeit nennt 43 Literaturangaben. Gefördert von: NHLBI NIH HHS.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study Objectives Several recent works predict incident cardiovascular disease (CVD) from routine sleep study signal data using artificial intelligence (AI) to better identify high-risk individuals and treatment candidates. However, these works often overlook sleep cohort dataset biases and lack simple comparative baselines to evaluate AI performance. Materials and Methods We examined relationships among age, sex, apnea hypopnea index (AHI), total sleep time, and percentage of slow wave sleep using two-way analysis of variance across three primary sleep cohorts: the Sleep Heart Health Study (SHHS), the Multi-Ethnic Study of Atherosclerosis, and the Osteoporotic Fractures in Men Study. Using SHHS data, we constructed two sets of Cox proportional hazard models to estimate CVD risk across six outcomes. One used age and sex as the only regressors, and another used non-sleep risk factors from the American Heart Association’s (AHA) Predicting Risk of Cardiovascular Disease EVENTs (PREVENT) risk calculator. We compared baseline performance against sleep AI models.

Results

Sleep cohorts consist primarily of older participants with clinically diagnosable sleep apnea based on AHI. Age-sex and AHA PREVENT Cox models predicted six CVD outcomes and closely matched results of AI models, including SleepFM, indicating sleep AI models may learn proxies for demographic features when estimating CVD outcomes.

Conclusions

Evaluating sleep AI models requires demonstrating clinically relevant improvements over traditional risk scores. Priorities to improve sleep AI capabilities include appropriately accounting for age as a risk factor, collecting more diverse data across age and healthy sleep, and leveraging explainability techniques to discover new features of sleep physiology that estimate CVD risk better than demographics alone.

Schlagwörter

Stichwörter der Autor:innen

AIAgingCardiovascular Disease RiskOSAPolysomnography

Autor:innen und Institutionen

  • The Windreich Department of Artificial Intelligence and Human Health, Icahn School of Medicine at Mount Sinai, New York, NY, USA.
  • Division of Pulmonary, Critical Care and Sleep Medicine, Department of Medicine, Icahn School of Medicine at Mount Sinai, New York, NY, USA.
  • The Charles Bronfman Institute for Personalized Medicine, Icahn School of Medicine at Mount Sinai, New York, NY, USA.
  • The Hasso Plattner Institute for Digital Health at Mount Sinai, Icahn School of Medicine at Mount Sinai, New York, NY, USA.

ORCID-Profile: Benjamin Fox, Ankit Parekh

Förderung

  • NHLBI NIH HHS

Zitieren

APA

Fox, B., Nadkarni, G. N., & Parekh, A. (2026). Are multimodal artificial intelligence predicting proxies for age and sex in sleep-based cardiovascular risk models?. SLEEP. https://doi.org/10.1093/sleep/zsag266

Vancouver

Fox B, Nadkarni GN, Parekh A. Are multimodal artificial intelligence predicting proxies for age and sex in sleep-based cardiovascular risk models?. SLEEP. 2026. doi:10.1093/sleep/zsag266

BibTeX
@article{Fox2026,
  author = {Fox, Benjamin and Nadkarni, Girish N and Parekh, Ankit},
  title = {Are multimodal artificial intelligence predicting proxies for age and sex in sleep-based cardiovascular risk models?},
  journal = {SLEEP},
  year = {2026},
  doi = {10.1093/sleep/zsag266}
}
Kostenlos starten →