AI Sleep Test Predicts Risk of Serious Diseases and Death Years in Advance

An artificial intelligence model developed by researchers in the United States can analyze routine sleep study data to identify individuals at elevated risk for cardiovascular disease, neurological disorders, and premature death years before symptoms appear, according to a study published in the journal Nature Communications.

Polysomnography, a complex sleep investigation commonly used to evaluate conditions like sleep apnea, contains significantly more health information than clinicians typically extract during standard daily practice. While doctors have historically focused on a narrow subset of metrics to measure sleep apnea severity, multidisciplinary researchers found that a single night of physiological recordings holds broader predictive power when analyzed with advanced computing tools.

AI Uncovers Risk Groups Missed by Traditional Metrics

The predictive model was created through the Discovery Accelerator, a partnership between the Cleveland Clinic in the United States and IBM aimed at accelerating life sciences research using artificial intelligence. Researchers analyzed comprehensive data from sleep studies and categorized patients into five distinct risk profiles. This nuanced stratification was not captured by the apnea-hypopnea index, which remains the primary clinical indicator used to measure the frequency of breathing interruptions during sleep.

According to the study, patients placed in the highest-risk group faced a probability of death over two times greater within a five-year window compared to those in the lowest-risk tier. Individuals in this vulnerable category also demonstrated a heightened susceptibility to subsequent cardiovascular and neurological conditions. Furthermore, the algorithm successfully identified these elevated risks in both male and female participants. By contrast, researchers noted that the traditional apnea-hypopnea index has historically exhibited stronger predictive value primarily in male patients.

Validating Sleep Physiology Beyond Standard Measures

To confirm the reliability of the findings, the research team validated the artificial intelligence model using an independent population cohort consisting of more than 6,000 participants drawn from the Sleep Heart Health Study.

Future Implications for Clinical Practice

Between one million and four million polysomnography tests are performed annually in the United States to diagnose sleep disturbances. Although these procedures capture continuous streams of data concerning brain activity, lung function, muscle movement, and cardiac rhythms, standard clinical workflows discard the vast majority of that physiological record after scoring for apnea.

The new analytical approach demonstrates that routine diagnostic tests already gathered in hospitals and clinics could be repurposed to flag hidden vulnerabilities. By extracting overlooked biomarkers from existing data sets, healthcare providers could eventually utilize automated stratification to schedule closer monitoring and implement personalized, early interventions for patients identified with high-risk physiological patterns.