Session Title: Genetics and Genomics of Rheumatic Disease II
Session Type: Abstract Submissions (ACR)
Recent advances in the identification of loci associated with susceptibility to complex disease have led to methods being developed that incorporate this information into genetic screening models to identify individuals at high risk of disease. Here we present the first risk prediction model including all 46 known genetic loci associated with rheumatoid arthritis (RA) in Caucasian populations.
Weighted genetic risk scores (wGRS) were created using odds ratios from 45 RA susceptibility SNPs and a HLA-DRB1 tag SNP or imputed HLA amino acids (HLA-DRB1 amino acids 11, 71, 74 and amino acids 9 at HLA-DPB1 and HLA-B). The wGRS were tested in 11,370 RA cases and 15,536 healthy controls of known genotype from the UK, USA, Sweden, Netherlands and Spain, all cases were Caucasian, over the age of 18 and satisfied 1987 ACR criteria for RA modified for genetic studies. The risk of developing RA was assessed using logistic regression by dividing the wGRS into quintiles. The ability of the wGRS to discriminate between cases and controls was assessed by receiver operator curve (ROC) analysis and discrimination improvement tests. As several of the RA susceptibility loci included in the wGRS were identified in the cohort used in this study, the wGRS were tested in an independent European cohort of 2206 RA cases and 1863 healthy controls for validation. Validation samples were from the Consortium of Rheumatology Researchers of North America (CORRONA) registry and the Informatics for Integrating Biology and the Bedside (I2B2) centre.
Individuals in the highest risk group showed significantly increased odds of developing anti-CCP positive RA compared to the lowest risk group (AUC 0.77, OR 18.00 95% CI 13.67-23.71). 4347 individuals were classed as high risk when all susceptibility factors were included in the model, interestingly this included 10.13% of our control population. The wGRS was validated in an independent cohort which showed similar results (AUC 0.78, OR 18.00 95% CI 13.67-23.71). The AUC was improved by replacing the HLA-DRB1 tag SNP with imputed HLA variation at HLA-DRB1, HLA-DPB1 and HLA-B (wGRStag AUC 0.70, wGRSfull AUC 0.77) (in CCP positive individuals). Integrated discrimination improvement tests showed an increase in sensitivity and specificity of 9% and net reclassification tests showed that including all variation in the HLA over a HLA tag SNP improves the probability of correctly identifying a high risk individual to 69.20% from 64.21% and showed an overall reclassification improvement of 58.54% between models. Comparing the full model (in CCP positive individuals) to a model containing only imputed variation at the HLA and gender showed that the addition of the susceptibility SNPs to the model only slightly improved the ROC AUC (0.74 to 0.77 respectively).
We have shown that in RA, even when using all known genetic susceptibility variants, prediction performance remains modest; whilst this is insufficiently accurate for general population screening, it may prove of more use in targeted studies. Our study has also highlighted the importance of including HLA variation in risk prediction models.
D. A. Pappas,
J. M. Kremer,
J. D. Greenberg,
R. M. Plenge,
« Back to 2013 ACR/ARHP Annual Meeting
ACR Meeting Abstracts - https://acrabstracts.org/abstract/a-weighted-genetic-risk-score-using-all-known-susceptibility-variants-to-predict-rheumatoid-arthritis-risk/