Assessing correlation of clustered mixed outcomes from a multivariate generalized linear mixed model

Area of Science:

• Statistics
• Biostatistics
• Multivariate Analysis

Background:

• The concordance correlation coefficient (CCC) traditionally measures agreement between two variables.
• Recent advancements generalized CCC for univariate generalized linear mixed models with exponential family distributions.
• Multivariate data, common in repeated measures and multiple methods, require new association indices.

Purpose of the Study:

• To propose novel correlation indices for multivariate generalized linear models.
• To address the need for measuring association in clustered mixed data, including within-observer and between-method correlations.
• To extend the concept of concordance correlation for joint modeling of count and continuous outcomes.

Main Methods:

• Development of intra-correlation, inter-correlation, and total correlation coefficients.
• Application within the framework of multivariate generalized linear models.
• Demonstration using simulation studies and a real-world case example.

Main Results:

• The proposed indices effectively measure various correlation aspects in multivariate settings.
• The methodology is shown to be robust through simulation studies.
• The indices provide a practical tool for analyzing complex data structures.

Conclusions:

• The new correlation coefficients are valuable extensions of the CCC for multivariate data.
• These indices facilitate a deeper understanding of associations in clustered and mixed outcomes.
• The proposed methods are applicable to joint modeling of diverse data types, as shown in the osteoarthritis example.