Application of bivariate mixed counting process models to genetic analysis of rheumatoid arthritis severity

We sought to i) identify putative genetic determinants of the severity of rheumatoid arthritis in the NARAC (North American Rheumatoid Arthritis Consortium) data, ii) assess whether known candidate genes for disease status are also associated with disease severity in those affected, and iii) determine whether heterogeneity among the severity phenotypes can be explained by genetic and/or host factors. These questions are addressed by developing bivariate mixed-counting process models for numbers of tender and swollen joints to evaluate genetic association of candidate polymorphisms, such as DRB1, and selected single-nucleotide polymorphisms in known candidate genes/regions for rheumatoid arthritis, including PTPN22, and those in the regions identified by a genome-wide linkage scan of disease severity using the dense Illumina single-nucleotide polymorphism panel. The counting process framework provides a flexible approach to account for the duration of rheumatoid arthritis, an attractive feature when modeling severity of a disease. Moreover, we found a gain in efficiency when using a bivariate compared to a univariate counting process model.