Concordance correlation coefficients estimated by variance components for longitudinal normal and Poisson data

The concordance correlation coefficient (CCC) is widely used to assess agreement between two observers for continuous responses. Further, the CCC is extended for measuring agreement with discrete data. This paper proposes a variance components (VC) approach that allows dependency between repeated measurements over time to assess intra-agreement for each observer and inter- and total agreement among multiple observers simultaneously under extended three-way generalized linear mixed-effects models (GLMMs) for longitudinal normal and Poisson data. Furthermore, we propose a weight matrix to compare with existing weight matrices. Simulation studies are conducted to compare the performance of the VC, generalized estimating equations and U-statistics approaches with different weight matrices for repeated measurements from longitudinal normal and Poisson data. Two applications, of myopia twin and of corticospinal diffusion tensor tractography studies, are used for illustration. In conclusion, the VC approach with consideration of the correlation structure of longitudinal repeated measurements gives satisfactory results with small mean square errors and nominal 95% coverage rates for all sample sizes.