Prediction model for rheumatoid arthritis: modelling 46 genetic risk variants with smoking

Abstract

Background Improved characterisation of the risk factors for rheumatoid arthritis raises the possibility that they could be combined to identify individuals at high risk of disease in whom preventive strategies can be evaluated. Our aim was to develop a prediction model capable of identifying such individuals with sufficient accuracy to enable the assessment of preventive treatments. MethodsOur prediction model combines odds ratios for 15 HLA-DRB1 alleles, 31 non-HLA single nucleotide polymorphisms (SNPs), and smoking status to classify an individual's risk of seropositive rheumatoid arthritis. Our novel modelling technique employs confidence intervals to classify disease risks using a computer-simulated population. We developed several models (HLA, HLA-10/20/31 SNP, HLA-smoking models) to evaluate the impact of different factors on prediction. The ability of each model to discriminate between rheumatoid arthritis and controls was evaluated in two European cohorts: the Wellcome Trust Case Control Consortium (WTCCC: 1542 cases, 1226 controls) and the UK Rheumatoid Arthritis Genetics Consortium (UKRAG: 2623 cases, 1503 controls).Findings HLA-DRB1 alleles conferred most prediction: the WTCCC HLA-only model classified 50% antibodies to citrullinated protein antigens (ACPA)-positive rheumatoid arthritis versus 17% controls as high risk and 60% controls versus 25% ACPA-positive rheumatoid arthritis as reduced risk. Adding smoking information improved prediction (p=0·00033); SNPs provided no significant benefits. The highest area under the curve was 0·81 (95% CI 0·78–0·85). Only a minority had substantially elevated risks of rheumatoid arthritis: 6·9% ACPA-positive cases and 0·31% controls in WTCCC had an odds ratio for rheumatoid arthritis of more than 20 when evaluated with the HLA-31 SNP model. InterpretationCombined information on HLA-DRB1 alleles and smoking provides informative risk prediction for rheumatoid arthritis. Since only a minority of individuals are at substantially elevated risks, modelling may be best focused in a-priori high-risk groups such as those with family histories of rheumatoid arthritis. Further work is needed to define risk factors with large effect sizes for incorporation within our modelling framework. FundingArthritis Research UK.