A scalable framework for population-level precision health: Insights from sleep disorders

Abstract

Assessing population-level risk patterns of sleep disorders is challenging. Many studies rely on hospital-based samples, which can be biased toward individuals with severe symptoms, or on polysomnography, which is costly and impractical for large-scale screening. To address this, we used SLEEPS, a machine learning algorithm that estimates risks of insomnia, obstructive sleep apnea (OSA), and comorbid insomnia and sleep apnea (COMISA). Since its online launch, SLEEPS has been used by over 40,000 individuals, producing a large web-based dataset collected through voluntary participation from public users. In this sample, total OSA risk, calculated as the sum of isolated OSA and COMISA risks, increased monotonically with age (τ∼0.453–0.465, p < 0.001), whereas total insomnia risk, calculated as the sum of isolated insomnia and COMISA risks, showed negligible associations (|τ|<0.1, p < 0.001). Decomposing total risks showed that only isolated OSA risk rose significantly with age. Comparison with hospital data showed that total insomnia risk was higher in the hospital sample than in the web sample, with differences up to 34.36% in women aged 45–49, while total OSA risk did not differ significantly. Sex-related differences were also evident: females had higher insomnia risk, whereas males had higher isolated OSA risk, consistent with known epidemiological patterns. These results demonstrate that SLEEPS provides a simple, scalable, and non-intrusive population-level risk assessment tool for insomnia, OSA, and COMISA in large-scale settings where polysomnography is not feasible. Moreover, this study highlights how digital data and machine learning can evaluate population-level risk patterns in other chronic diseases where comprehensive diagnostics are impractical. © 2026 Elsevier B.V.