A Bayesian Method to Reduce Selection Bias in Genetic Association Studies of CKD Progression

Background: The rate of progression of CKD has high inter-patient variability. Genetic differences are responsible for some of this variation, but GWAS results have been inconsistent. This may be due to selection bias arising when GWAS participants are selected based on kidney function. Such bias is apparent in the results of existing studies. Current methods to address this bias are predicted to fail in the setting of CKD due to high genetic correlation between incident CKD and rate of CKD progression. Methods: We adapted a Mendelian Randomization method (MR-Horse) for unbiased estimation of SNP associations with CKD progression in the setting of correlated SNP effects on disease incidence and progression (the hypothesized genetic architecture of CKD). We applied this method and others to simulated GWAS of traits with different genetic architectures. We then applied these methods to GWAS results from the CKDGen Consortium, and evaluated their effects on SNP-progression estimates at loci with known significant effects on progression and at loci with likely biased associations. Results: Our method reduced the bias of simulated SNP-progression effect estimates, including in situations with high genetic correlation where other methods failed (Fig 1). Power and type 1 error rate were comparable to other methods. In CKDGen data (Fig 2), our method attenuated associations at loci with likely biased associations with CKD progression, whilst preserving associations at loci with likely real direct effects on progression. Conclusion: GWAS of CKD progression are affected by selection bias, and this can be reduced by an adaptation of the MR-Horse method. Funding: Private Foundation Support