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

Area of Science:

• Digital Health
• Machine Learning
• Sleep Medicine

Background:

• Assessing population-level sleep disorder risk is difficult due to biased samples and costly diagnostics.
• Traditional methods like hospital-based sampling and polysomnography have limitations for large-scale screening.

Purpose of the Study:

• To introduce SLEEPS, a machine learning algorithm for estimating population-level risks of insomnia, obstructive sleep apnea (OSA), and comorbid insomnia and sleep apnea (COMISA).
• To evaluate the performance and utility of SLEEPS using a large, web-based dataset.

Main Methods:

• Utilized SLEEPS, a machine learning algorithm, to analyze data from over 40,000 voluntary public users.
• Calculated total and isolated risks for insomnia and OSA.
• Compared web-based data with hospital-based samples to assess potential biases and differences.

Main Results:

• Total obstructive sleep apnea (OSA) risk increased significantly with age (τ∼0.453-0.465, p < 0.001).
• Total insomnia risk showed negligible association with age (|τ|<0.1, p < 0.001).
• Insomnia risk was higher in hospital samples compared to web samples, particularly in women, while OSA risk did not differ significantly.

Conclusions:

• SLEEPS offers a scalable, non-intrusive tool for population-level risk assessment of insomnia and OSA where polysomnography is impractical.
• This study demonstrates the potential of digital data and machine learning for evaluating chronic disease risk patterns in large populations.