A joint modeling and estimation method for multivariate longitudinal data with mixed types of responses to analyze...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Wearable Technology Data Analysis

Background:

• Analyzing multivariate longitudinal data with mixed response types (continuous, proportion, count, binary) presents statistical challenges.
• Existing models may not adequately capture the complex associations between different types of longitudinal outcomes.
• Accurate analysis of data from wearable devices, like accelerometers, is crucial for understanding health behaviors.

Purpose of the Study:

• To propose a novel mixed-effects model for the joint analysis of multivariate longitudinal data with diverse response variables.
• To effectively model the associations between different response types using the correlation of random effects.
• To develop an efficient estimation algorithm and apply the method to real-world physical activity data.

Main Methods:

• A mixed-effects model is proposed to jointly analyze continuous, proportion, count, and binary longitudinal data.
• Associations are modeled via the correlation of random effects, with quasi-likelihood approximation for nonlinear variables.
• The model is transformed into a multivariate linear mixed model framework, fitted using an EM algorithm extension.

Main Results:

• The proposed method successfully analyzes multivariate longitudinal data with mixed response types.
• The model demonstrates the ability to capture associations between different variables through random effects.
• Application to physical activity data from wearable accelerometers provides valuable insights into movement and energy expenditure.

Conclusions:

• The developed mixed-effects model provides a flexible and efficient framework for joint analysis of complex longitudinal data.
• The method is robust and applicable to diverse datasets, including those from wearable health monitoring devices.
• Simulation studies and real-world data application confirm the model's validity and utility in biostatistical research.