Methodological Considerations to Enhance the Relevance of Causal Observational Research in the Field of Kidney Disease

The recent study published in JASN by Cheng et al.1 examines the impact of initiating statin therapy in patients with kidney failure. However, several aspects of the study warrant closer examination to ensure that the findings meaningfully inform clinical practice. First, focusing on incident statin users limits the study's relevance because most patients already use statins or have cardiovascular disease at the time of kidney failure. Of 26,933 individual patients, only 3019 were eligible, and 17% of them initiated statins. A more clinically meaningful question would be whether patients with kidney failure should continue or stop statins, similar to previous studies that have explored the withdrawal of renin-angiotensin system inhibitors in advanced kidney disease.2 In addition, the inability to distinguish between patients on dialysis and nondialysis patients is problematic because these groups may respond differently to statins. It is also unclear whether kidney transplant recipients were included and how transplantation during follow-up was handled. Second, several analytical choices violate the principles of target trial emulation, even though the study reports to implement this framework. A key aspect of this approach is that three aspects need to be aligned at one moment in time: meeting eligibility criteria, assigning treatment strategies, and commencing follow-up.3 Misalignment of these elements introduces biases.3 Here, the authors' exclusion of patients who experienced an outcome during the first year of follow-up introduces selection bias. In addition, the requirement to have a follow-up visit after baseline introduces immortal time because patients must survive to the first follow-up to be included. These issues, which may have contributed to the low incidence of outcomes observed in the study, impede accurate estimation of absolute risks. Finally, by censoring patients upon death and applying censoring weights, the findings reflect the effect of statin use in an unrealistic hypothetical scenario where no individuals ever die, which, in causal terms, corresponds to the controlled direct effect.4 A causal estimand that would be more relevant for decision making would be the total effect of statin use on cardiovascular disease, which is the composite of the direct effect and the effect mediated by the competing risk of death. We believe that addressing these methodological considerations would enhance the robustness of the findings and their relevance to patient care. Future research may benefit from a design that better reflects real-world clinical practice, aligns with target trial emulation principles, and targets clinically meaningful estimands when dealing with competing risks.