Arthritis & Rheumatism, Volume 62,
November 2010 Abstract Supplement
Abstracts of the American College of
Rheumatology/Association of Rheumatology Health Professionals
Annual Scientific Meeting
Atlanta, Georgia November 6-11, 2010.
Combining Genetic and Environmental Risk Factors To Model RA Susceptibility.
Chibnik3, Lori B., Ding7, Bo, Keenan4, Brendan T., Liao2, Katherine P., Costenbader1, Karen H., Klareskog6, Lars, Alfredsson7, Lars
Brigham & Women, Boston, MA
Brigham & Women's Hosp, Boston, MA
Brigham and Women's Hospital, Boston, MA
Brigham and Women's Hospital
Brigham and Womens Hosp, Boston, MA
Karolinska University Hospital, Stockholm, Sweden
Karolinska University Hospital
Cumulative genetic risk scores (GRSs) have shown promise in modeling rheumatoid arthritis (RA) susceptibility. In addition, many gene-environment interactions (GEIs) have been significant factors in RA risk. We combine a GRS and GEIs to develop the best model for RA risk.
We studied models of RA risk in two cohorts, the U.S. Nurses' Health Studies (NHS) and the Swedish Epidemiologic Investigation of RA (EIRA). We created a weighted GRS using 31 non-HLA single nucleotide polymorphisms at validated RA risk loci, where the weight for each risk allele is the log of the odds ratio based on published GWAS or meta-analysis. In addition, we genotyped HLA-DRB1 locus with subjects coded as having 0, 1 or 2 copies of shared epitope (SE) alleles. Based on previously found significant interactions, we included two additional polymorphisms, GSTT1-null and HMOX1. Smoking was the main environmental factor for GEI (dichotomized as <= 10 pack-years vs. > 10 pack-years). Parallel analyses were performed among (1) 371 Caucasian seropositive (CCP+ and/or RF+) cases and 551 controls from NHS and (2) 987 Caucasian ACPA positive cases and 958 controls from EIRA to develop the best model in each dataset. We began with a base model including year of birth and smoking (plus sex and region of Sweden in EIRA). Hierarchical models adding the most significant factors were compared to the previous and base models using pseudo R2 for parsimony and area under the ROC curve (AUC) and the Integrated Discrimination Improvement (IDI, which quantifies the overall improvement in sensitivity and specificity) for discrimination.
The mean (SD) age of diagnosis of RA was 57 (10) in NHS and 50 (12) in EIRA. For NHS, the base model produced an AUC of 0.578. Adding the GRS and HLA to the model increased the AUC to 0.669. The IDI between the 2 models was 0.07 (0.050.08), indicating a significant improvement (null IDI = 0). Our best performing model contained the GRS and GEIs between smoking and HLA, GSTT1-null and HMOX1 and had an AUC of 0.700 and IDI of 0.10 (0.070.12) compared to the base model and 0.03 (0.020.04) as compared to the model with GRS and HLA. For EIRA, the base model produced an AUC of 0.630. Adding the GRS and HLA to the model increased the AUC to 0.733. The IDI between the 2 models was 0.11 (0.100.13). Adding the GEIs between smoking and HLA did not significantly improve the model (AUC = 0.734, IDI = 0.001 [-0.0030.003]), compared to the model with GRS, clinical factors and HLA. Inclusion of other covariates did not improve the models in either cohort.
Inclusion of GEIs significantly improves the discriminative ability of models predicting RA risk in NHS but not EIRA. These conflicting results emphasize the importance of developing models separately in different populations. However, further work to discover genetic, environmental and GEI factors is needed before these models are used in clinical settings.
Table. Summary Measures for Selected Models in the US Nurses' Health Study and Swedish Epidemiologic Investigation in RA
To cite this abstract, please use the following information:
Chibnik, Lori B., Ding, Bo, Keenan, Brendan T., Liao, Katherine P., Costenbader, Karen H., Klareskog, Lars, et al; Combining Genetic and Environmental Risk Factors To Model RA Susceptibility. [abstract]. Arthritis Rheum 2010;62 Suppl 10 :1588