Impact of selection bias on the association of kidney function polygenic scores with CKD progression in populations with CKD

Abstract Background and Hypothesis The associations of polygenic scores (PGS) for eGFR, with clinical outcomes in prevalent CKD populations have not been well evaluated. Collider bias may distort associations of exposures (including PGS) with outcomes in populations with prevalent disease Methods Analyses included European ancestry participants from UK Biobank and three CKD cohorts (SHARP, CRIC and EMPA-KIDNEY) without ESKD at baseline. The exposure was a PGS for low eGFR derived from GWAS of eGFR in the FinnGen cohort. The primary outcomes were eGFR slope (mls/min/1.73m2/year) and kidney disease progression (KDP; defined as first occurrence of incident dialysis, transplantation, sustained eGFR decline by ≥ 40% or sustained eGFR  Results 107 536 participants in UKB and 6 846 participants from CKD cohorts were included in analyses. In UKB higher PGS was associated with lower baseline eGFR (−5.09 [−5.31; −4.86] ml/min/1.73m2 for those in the top 10% of the PGS versus the rest), with greater risk of KDP (HR 1.36 [1.20–1.55]) and faster eGFR decline (-0.12 [-0.14;-0.09] ml/min/1.73m2/year). However, in CKD cohorts, higher PGS was not associated with baseline eGFR (0.15 [−0.73; 1.03] ml/min/1.73m2) but was associated with lower risk of KDP (HR 0.76 [0.66–0.89]), with a directionally consistent (though non-significant) effect on eGFR decline (+0.19 [-0.01; 0.39] ml/min/1.73m2/year). Conclusions A PGS for low eGFR is associated with faster eGFR decline and higher risk of KDP in a population unselected for CKD, but when applied in populations with CKD, non-causal associations are introduced which artificially reverse the associations seen. Applying PGS in populations selected for prevalent disease may lead to qualitatively misleading conclusions on the direction of associations with clinical outcomes.