A Novel Sparse Linear Mixed Model for Multi-Source Mixed-Frequency Data Fusion in Telemedicine

Area of Science:

• Digital Health
• Biostatistics
• Cardiology

Background:

• Digital health and telemonitoring generate vast, complex datasets.
• Existing models struggle with multi-source, mixed-frequency health data.
• Automated prediction of Disease Severity Indicators (DSIs) for sleep disorders is lacking.

Purpose of the Study:

• To develop a rigorous prediction model for DSIs from multi-source, mixed-frequency data.
• To address challenges in high-dimensional data for sleep disorder telemonitoring.
• To enable automated monitoring and diagnosis of sleep disorders.

Main Methods:

• Proposed a sparse linear mixed model using modified Cholesky decomposition and group lasso penalties.
• Developed a novel Expectation Maximization (EM) algorithm integrated with Majorization Maximization (MM) for model estimation.
• Applied the method to the SHHS dataset for sleep disorder telemonitoring and diagnosis.

Main Results:

• Identified significant feature groups consistent with existing sleep disorder research.
• The proposed method demonstrated superior prediction accuracy compared to benchmark approaches.
• Successfully applied the model to real-world telemonitoring data for sleep disorder diagnosis.

Conclusions:

• The developed sparse linear mixed model effectively predicts DSIs from complex health data.
• This approach enhances automated sleep disorder monitoring and diagnosis.
• The findings support the use of advanced statistical modeling in digital health applications.