A multidisciplinary research team has developed an artificial intelligence model that extracts hidden physiological signals from routine overnight sleep studies, revealing long-term health risks that conventional clinical metrics miss. Published in Nature Communications, the study demonstrates how foundation models can analyze complete polysomnography data to stratify patients into five distinct prognostic categories, advancing predictive health analytics. Historically, clinicians have distilled sleep studies into a limited set of summary measures, primarily the apnea-hypopnea index, to grade sleep apnea severity. The new AI framework processes rich, multi-system data encompassing brain activity, respiratory patterns, cardiovascular function, and muscle signals. By learning from the full physiological profile of each overnight study, the model identifies latent biomarkers associated with cardiovascular disease, cognitive decline, and all-cause mortality. Clinical validation revealed sharply divergent outcomes across the identified risk groups. Patients in the highest-risk tier exhibited a two-fold increase in five-year mortality compared to those in the lowest tier, a critical prognostic distinction entirely absent from standard severity scores. Notably, the AI maintained consistent predictive accuracy across both male and female demographics, addressing historical performance gaps in traditional metrics. Independent confirmation in a nationwide patient cohort further substantiated the model robustness. The research was driven by a cross-institutional partnership anchored by the Cleveland Clinic and IBM Discovery Accelerator, a decade-long initiative advancing life sciences innovation through artificial intelligence and quantum computing. Lead investigators from the Cleveland Clinic, University of Washington, and Yale School of Medicine emphasized that the findings underscore a systemic opportunity to repurpose existing clinical data. Routine medical examinations currently underutilize the dense physiological information captured during standard diagnostics. By applying advanced machine learning, healthcare providers can transition from reactive diagnosis to proactive risk stratification. Sleep disorders affect nearly 70 million Americans and serve as a foundational indicator of broader systemic health. The AI-driven approach promises to transform sleep medicine by integrating overnight data into comprehensive cardiovascular and neurological prognostic frameworks. This enables earlier clinical intervention, more personalized treatment pathways, and more efficient medical resource allocation. Researchers now plan to validate the model across diverse populations while expanding collaborations among academic institutions, industry partners, and clinical societies. The study marks a significant milestone in clinical AI, proving that foundation models can successfully decode complex medical signals to generate actionable health intelligence. As healthcare systems prioritize predictive analytics, mining latent data from widely available tests establishes a scalable blueprint for next-generation diagnostic innovation.