A stochastic regression model for general trend analysis of longitudinal continuous data

Area of Science:

• Statistics
• Biostatistics
• Time Series Analysis

Background:

• Continuous panel data requires models that account for time-varying covariates.
• Existing discrete time models may not be optimal for unequally spaced data.
• Understanding the dynamics of continuous responses influenced by covariates is crucial in longitudinal studies.

Purpose of the Study:

• To develop a predictive continuous time model for panel data.
• To assess the impact of time-varying covariates on a continuous response.
• To provide a robust method for analyzing longitudinal data with potential missing observations.

Main Methods:

• Reparameterization of Ornstein-Uhlenbeck processes using equilibrium mean and drift parameters.
• Modeling the equilibrium mean as a linear predictor of covariates.
• Comparison of maximum likelihood and quasi-likelihood estimation approaches via simulation.
• Application to diastolic blood pressure data from a cardiovascular disease study.

Main Results:

• The proposed continuous time model effectively assesses covariate effects on fluctuating responses.
• The quasi-likelihood approach demonstrated high efficiency and robust variance estimation compared to maximum likelihood.
• The model naturally accommodates missing observations in longitudinal datasets.

Conclusions:

• The developed continuous time model offers a flexible and robust framework for analyzing panel data with time-varying covariates.
• The quasi-likelihood method is a practical and efficient choice for parameter estimation in this context.
• This approach is well-suited for applications in fields like cardiovascular disease research where longitudinal data is common.