Phenotypic cluster analysis of Juvenile Idiopathic Arthritis: relationship to ILAR classification criteria

Background/Purpose:

The ILAR classification of Juvenile Idiopathic Arthritis (JIA) delineated categories of childhood chronic arthritis based on predominant clinical and laboratory features.   As consensus criteria, they may not reflect clinically significant disease patterns that exist among patients. Phenotypic analysis of patient characteristics may provide a more optimal way to group patients.  The aim of this study was to identify distinct clusters of JIA patients based on demographic, clinical, laboratory, and imaging features.

Methods:

The source population was subjects with JIA enrolled in the CARRA Registry from May 2010 to April 2012, representing 60 centers throughout the US.  Demographic, clinical, laboratory, and imaging data were gathered through patient reporting, physician assessment, and chart review.  K-means clustering analysis was performed to identify distinct clusters of patients based on major features.  Weighted gap statistics were used to estimate the number of clusters present in the dataset.  Association between cluster assignment and physician-assigned ILAR JIA category was determined using Fisher’s exact test.

Results:

A total of 4810 subjects with JIA were enrolled in the CARRA registry during the enrollment period.  3872 patients had data for all items used in the cluster analysis and were therefore included.  10 distinct clusters were identified and are shown in Table 1. For each variable used in the analysis, at least one cluster was significantly different (p<0.001) from all other clusters. Fisher's exact test revealed that cluster assignment was highly correlated (p<0.001) to physician-assigned ILAR JIA category (see Table 2).

Conclusion:

Children with JIA have distinct phenotypical patterns of demographics, and clinical, laboratory, and imaging findings. However, patient groupings based on these patterns may not be of more utility than the currently available, consensus-driven, categorizations. Further registry data that provides patient outcome data, which is actively being collected, will be of use to determine whether phenotypic clustering has prognostic value.

Table 1: Distribution of demographic, clinical, laboratory and imaging variables among clusters in JIA patients

a: ever or in past

Table 2: Comparison of cluster assignment to physician-assigned ILAR JIA category