Agreement concordance correlation coefficient

Question

I need to compute the agreement between 3 readers or at least 2 pairs of readers using a concordance correlation coefficient for correlated non longitudinal data. 3 reviewers measure about 20 lesions. 1 to 4 lesions per patients. To my knowledge, JL Carrasco's R package "CCCRM" and L Lin & Y Yu's R package "Agreement" do not explicitly give an example of non-repeated non-longitudinal measurements where the clusters (the patients) contain a variable number of measurements. There is no missing values. No unbalanced design (each reader measures any of the 20 lesions).

With non longitudinal data, I exclude autoregressive correlation structure and fixed effect for time.

Any suggestion or reference paper? For example: best approach for the variance estimates: U-statistics, GEE, or linear mixed models? Try Lin & Yu's GGE based unified approach and some post-hoc correction for the within-patient correlation?

Answer 1

For other readers who need a intraclass correlation coefficient (ICC) or a concordance correlation coefficient (CCC) for intra-rater agreement, inter-rater agreement or absolute agreement in the presence of repeated measurements because of clusters (several locations or lesions per patient) where the longitudinal (based on fixed effect for time) approach does not work: See CC Chen & H Barnhart's paper: "Assessing agreement with intraclass correlation coefficient and concordance correlation coefficient for data with repeated measures", Computational Statistics and Data Analysis 60 (2013) 132–145. The use of random effect model for the time or location allows the computation of both ICC and CCC.