Semiparametric models and inference for biomedical time series with extra-variation

Area of Science:

• Biostatistics
• Time Series Analysis
• Biomedical Data Analysis

Background:

• Biomedical trials generate time-series data for individual subjects.
• Existing random effects models use sample periodograms to explain inter-individual variations.
• Standard asymptotic theory is used for individual series, with random effects for inter-individual differences.

Purpose of the Study:

• To extend existing random effects models for biomedical time-series data.
• To enable nonparametric specification of spectral behavior.
• To address computational challenges in random effects modeling.

Main Methods:

• Developed a model specifying population spectrum nonparametrically via a dynamic system.
• Assumed individual process spectra have random effect perturbations from the population norm.
• Utilized standard Markov Chain Monte Carlo (MCMC) algorithms for inference.

Main Results:

• Simulation studies confirmed effective inference using standard MCMC algorithms.
• The model successfully revealed scientifically important temporal structures within and between individual processes.
• Applications to biomedical data demonstrated the model's utility.

Conclusions:

• The enhanced random effects model provides a flexible framework for analyzing biomedical time-series data.
• Nonparametric spectral analysis and improved computation offer deeper insights into individual and population-level temporal dynamics.
• The model is effective for uncovering significant biological patterns in temporal data.