Mixture analysis of heterogeneous physical activity outcomes

Area of Science:

• Biostatistics
• Public Health
• Health Outcomes Research

Background:

• Physical activity (PA) offers significant health benefits.
• PA data frequently presents a semi-continuous distribution with excess zeros and right-skewed positive values.
• Clustered PA data often exhibits heterogeneity, requiring specialized analytical approaches.

Purpose of the Study:

• To propose a novel two-part mixture regression model with random effects for analyzing clustered physical activity data.
• To characterize the heterogeneity inherent in physical activity outcome data.
• To provide a statistical framework for understanding the complex nature of physical activity patterns.

Main Methods:

• A two-part mixture model was developed, incorporating a logistic mixed regression for the binary component (PA participation) and a gamma mixture regression for the continuous component (PA intensity).
• Random effects were integrated into both model components to address the correlation within clustered observations.
• Model fitting and inference were conducted using the Gaussian quadrature technique, implemented via SAS PROC NLMIXED.

Main Results:

• The proposed mixture model effectively handles the semi-continuous, right-skewed nature of physical activity data.
• The methodology was motivated by and applied to a study of physical activity in patients with chronic obstructive pulmonary disease (COPD).
• The statistical approach allows for robust model fitting and inference on complex PA data structures.

Conclusions:

• The developed mixture analysis is highly effective for understanding physical activity patterns.
• This approach enables the distinct identification of factors influencing physical activity participation versus intensity.
• The findings highlight the utility of mixture models for analyzing PA in diverse patient subgroups, particularly those with chronic conditions like COPD.