Bayesian structural time series for biomedical sensor data: A flexible modeling framework for evaluating...

Area of Science:

• Biomedical Informatics
• Personalized Medicine
• Statistical Modeling

Background:

• Mobile-health technology and biomedical sensors offer unprecedented data for personalized medicine.
• Interpreting complex biomedical sensor data requires advanced statistical models accounting for covariates and temporal/spatial properties.
• Existing methods may not adequately capture the causal impact of health interventions in rich, complex datasets.

Purpose of the Study:

• To demonstrate the application of the Bayesian structural time series (BSTs) framework for analyzing biomedical sensor data.
• To introduce MhealthCI, a software tool designed to process and prepare diverse biomedical data for BSTs analysis.
• To evaluate the efficacy of BSTs in assessing intervention impact with varying data complexity and covariate structures.

Main Methods:

• Utilized the Bayesian structural time series framework, adapted from economics, for analyzing biomedical sensor data.
• Developed and applied a biomedical data processing tool, MhealthCI, to uniformly handle diverse sensor inputs.
• Tested the framework on datasets including diabetes (exercise intervention) and behavioral data (spatial covariates).

Main Results:

• The BSTs framework accurately assesses intervention significance by correcting for covariates.
• MhealthCI successfully processed varied biomedical data, enabling robust analysis.
• Demonstrated the framework's ability to evaluate interventions like exercise on blood glucose and integrate complex spatial covariates.

Conclusions:

• The Bayesian structural time series framework is robust for analyzing biomedical sensor data.
• MhealthCI facilitates the application of BSTs for personalized medicine and intervention evaluation.
• This approach enhances the assessment of individualized intervention efficacy using mobile health data.