We selected individuals from Vanderbilt University’s BioVU repository with a diagnosis of rheumatoid arthritis48 using a previously validated algorithm for identification of RA cases with a reported positive predictive value of 0.94 and sensitivity of 0.87, as previously described52. This trained machine learning classifier was applied to records with at least one International Classification of Diseases, 9th edition code for rheumatoid arthritis to identify true RA cases. RA positive individuals identified by this algorithm were genotyped on two platforms: 833 using the Illumina OmniExpress + Exome chip and 1408 using the Illumina Omni 2.5 BeadChip. A total of 2650 samples from the Illumina Genotype Control set genotyped on Illumina HumanMap550v1/v3 were used for controls. We used the following QC thresholds: sample call rate > 0.98, SNP call rate > 0.99, MAF > 0.05, HWE p-value > 10−3. Imputation was performed using IMPUTE2 with the 1000 Genomes phase 1 v3 European samples as the reference panel, phasing was done with SHAPEIT, and SNPs with imputation quality score (“INFO”) > 0.50 were retained. To replicate the PrediXcan RA findings that meet genome-wide
