Simulation of longitudinal exposure data with variance-covariance structures based on mixed models.

Area of Science:

• Environmental Health Sciences
• Biostatistics
• Computational Biology

Background:

• Longitudinal data are crucial for exposure and risk assessments, particularly for pollutants with long biological half-lives.
• Collecting large longitudinal human exposure datasets is often difficult and costly.
• Chronic exposure to environmental pollutants raises concerns for human health effects.

Purpose of the Study:

• To introduce a novel simulation method for generating longitudinal human exposure data.
• To enable flexible generation of data with varying numbers of subjects and days.
• To provide a cost-effective alternative to empirical data collection.

Main Methods:

• Utilized mixed models to characterize variance-covariance structures of existing longitudinal data.
• Generated simulation data based on estimated model parameters.
• Employed criteria including mean, standard deviation, variance components, and autocorrelation coefficients to ensure statistical similarity.

Main Results:

• The simulation method successfully generated longitudinal exposure data meeting predefined statistical criteria.
• Simulated data accurately retained inter- and intra-subject variance proportions and autocorrelation patterns.
• The new method demonstrated superior performance in preserving overall data distribution compared to existing algorithms.

Conclusions:

• The developed simulation method effectively generates statistically similar longitudinal human exposure data.
• This approach offers a flexible and efficient tool for exposure and risk assessment studies.
• The method enhances the ability to simulate continuous observed variables and accommodates diverse user requirements.