Within-sibling attenuation of polygenic risk score accuracy: investigating the effects of principal component analysis, LD score regression, and mixed model association in the UK Biobank.

A central challenge in polygenic risk prediction is measuring and controlling for confounding due to population stratification. Standard approaches include adjusting for leading principal components (PCs) of genetic variation and using linear mixed models in genome-wide association studies (GWAS). Evidence of adequate control is typically inferred from reductions in the linkage disequilibrium score regression (LDSC) intercept and the preservation of polygenic risk score (PRS) performance in within-sibling settings. Here, we investigate how the number of PCs included during GWAS/PRS construction relates to the loss of predictive performance when moving from population-based analyses to that of discordant sibling pairs (within-sibling attenuation). This design detects confounding arising from environmental and genetic background effects correlated with population structure. Analyses were performed in the self-identified White subset of the UK Biobank (UKB) for coronary artery disease, type 2 diabetes, breast cancer, and prostate cancer, with educational attainment included as a comparison trait as it is known to exhibit substantial within-sibling attenuation. We find that increasing the number of PCs does not consistently reduce within-sibling attenuation across traits. Furthermore, reductions in attenuation do not closely correspond to decreases in the LDSC intercept, and mixed model-based methods offer little additional improvement in prediction or attenuation. Overall, our results suggest that confounding targeted by PC-adjustment and linear mixed models is either minimal in the UKB, or remains inadequately captured by current methods, reflecting limitations as to how much standard population structure correction improves the causal validity of PRS in the UKB.