Authors' Reply: Association between CKD and Kidney Stones: A Reliable Relationship?

We thank Zheng and Zeng1 for their insightful comments regarding our recent Mendelian randomization (MR)–based study, published in JASN, to re-examine the relationship between CKD and kidney stones.2 We particularly agree with the assessment that the intriguing results offer new directions for future etiological research and clinical interventions for this population of patients. In response to their methodologic observation and suggestions, we would like to stress that we used a literature-based approach to exclude single-nucleotide polymorphisms (SNPs) with potential pleiotropy rather than a purely statistical method. This approach is more specific and avoids overgeneralizing pleiotropy. For example, while fat mass and obesity associated gene-related SNPs are statistically associated with both obesity and diabetes, it is primarily considered an obesity-related gene.3 Removing these SNPs in MR analysis could potentially underestimate the effect of obesity on diabetes. After literature reviewing of the related genes of SNPs selected as instrumental variables, we only excluded UMOD-related SNPs in the MR analysis focused on CKD since its association with CKD is well-established.4 In addition, the UMOD locus may be associated with kidney stones independently of kidney function, despite inconsistent findings in observational studies on urinary uromodulin levels in stone formers.4 Reassuringly, our sensitivity analysis (Supplemental Figures 7 and 8) confirmed consistent results, even after excluding UMOD-related SNPs in the MR analysis of the effect of CKD on kidney stone risk.2 We entirely agree that MR statistical analysis methods are a rapidly evolving field. Nevertheless, multivariable MR remains a well-established, valid, versatile, and thus widely used approach.5 Importantly, the instrumental variables for kidney stones and CKD used in the multivariable MR are generally not associated with cardiometabolic conditions, ensuring a more accurate estimation of their effect sizes. Nevertheless, our group remains open to using these newer methods in future studies. Most importantly, regardless of the modeling method used, MR is not a perfect tool for establishing causality. To partially address this concern, we included additional observational analyses in this article, which showed consistent results and serve to help validate our main findings.