Statins at Dialysis Transition—Supportive but Not Sufficient

Despite the high cardiovascular (CV) disease burden and disordered lipid metabolism characteristic among individuals with advanced kidney disease, treatment with 3-hydroxy-3-methyl-glutaryl–coenzyme A (HMG-CoA) reductase inhibitors (statins) has yielded conflicting results on CV outcomes. Investigations such as the SHARP (Study of Heart and Renal Protection) trial have demonstrated statin-mediated reduction of CV morbidity in patients with chronic kidney disease (CKD).¹ However, dialysis subgroup analysis of SHARP and other randomized trials investigating statins for primary prevention of CV events in end-stage kidney disease (ESKD), such as the 4D study (Die Deutsche Diabetes Dialyse Studie) and AURORA study (A Study to Evaluate the Use of Rosuvastatin in Subjects on Regular Hemodialysis: An Assessment of Survival and Cardiovascular Events), have found no benefit of statin therapy in this population.^2,3 Despite the widespread prescription of statins and the high CV burden in CKD, the disparate findings before and after dialysis initiation have led to a paucity of reliable information regarding the use of statins around the initiation of dialysis. The 2013 KDIGO (Kidney Disease: Improving Global Outcomes) clinical practice guidelines recommended statin therapy for all patients with CKD who are at least 50 years of age and in younger patients with concomitant risk factors for coronary artery disease.⁴ While the guidelines recommend against starting statin therapy de novo in ESKD, KDIGO cited expert recommendation to continue preexisting statin therapy after the initiation of dialysis.⁴

Streja et al⁵ sought to fill the void of evidence for this recommendation. Their study investigates the association of statin discontinuation following dialysis initiation with all-cause and CV mortality in patients within the Veteran Affairs (VA) system from October 2007 through March 2014. The primary analysis separates veterans receiving statins for 1 year preceding dialysis initiation into statin discontinuers (<6 months of statin therapy in the first year of dialysis) with the remainder considered statin continuers. Streja and colleagues report a significant decrease in both all-cause and CV mortality in the statin continuer group compared with statin discontinuers, which remains consistent in numerous subgroup and sensitivity analyses.

In the absence of randomized clinical trials designed to address the question, under certain assumptions, well-defined and executed observational analyses can shed light on the nature of the associations. It is imperative to explore these assumptions and design features to gain a better sense of the reliability of presented conclusions. Streja and colleagues leveraged a large and rich real-world database from the VA system to address the question.

A critical assumption in the interpretation of their and any observational study results is the absence of unmeasured confounding. Because it is unmeasured, it cannot be tested. To their credit, the authors performed an important sensitivity analysis based on E-values, which quantify how strong the unmeasured confounding would have to be to “explain away” the observed associations. The authors report that “for an unmeasured characteristic to render the described association between statin continuation and mortality nonsignificant, it would have to show an adjusted risk ratio of at least 1.82 (or 1.63 to explain away the lower confidence limit) with both the exposure and the outcome separately.”⁵ They point out that this is a stronger association than that present for any of the measured characteristics. On the other hand, one might argue that such an association could be plausible with multiple confounders, for example, those related to health deterioration.

The primary analysis relied on Cox proportional hazards models adjusted for demographic factors and CV comorbidities. This was supplemented by sensitivity analyses based on propensity score adjustment and matching. The results of all of these analyses are consistent, which is reassuring. Still, a number of factors aligned with statin discontinuation were not adjusted for in the primary analysis or in the propensity model. These include angiotensin-converting enzyme discontinuation, the presence of arteriovenous fistula, and the total number of prescription medications. While additional adjustment for each factor individually in the supplementary material did not significantly alter the association with all-cause mortality, an aggregate effect cannot be fully ruled out.

Another important consideration is the extent to which the analyzed sample represents the population of interest. The authors report only a small amount of missing data (<0.4%) addressed by imputation. However, the study excluded a significant number of patients prior to the primary analysis due to missing data. If such data were not missing at random (related to unobserved characteristics), this approach may result in bias. Two excluded cohorts of interest in this study are those with missing prescription information and those dying within the first year of dialysis. For example, patients without nephrology care prior to dialysis initiation may be more likely to have missing prescription data. As demonstrated by the current study and others, predialysis nephrology care is associated with outcomes after dialysis initiation.⁶

Another challenging task the authors had to address was defining an appropriate duration of statin exposure before and after initiation of dialysis. While changes in lipid profile occur within weeks, the follow-up for clinical outcomes in the SHARP, 4D, and AURORA studies ranged from 3.2 to 4.9 years.^1-3 While the requisite duration of statin exposure needed to produce CV protection in CKD and ESKD is unclear, extension of the “run-in” period of statin therapy in the study by Streja and colleagues beyond 1 year may better facilitate comparison with these prior randomized trials.

The risk of mortality during the first year of dialysis is exceptionally high, with the risk during the first 2 weeks reaching nearly 3-fold that of the mortality risk during months 12 through 24.⁷ The study by Streja and colleagues excluded these patients, likely to limit statin discontinuation during the transition period in the setting of limited life expectancy or to avoid the issue of immortal time bias. Alternative approaches might have been considered to retain this important patient group. The first concern is difficult to eliminate entirely, but detailed propensity score matching could have reduced it somewhat. Immortal time bias could have been addressed with the use of a Cox model with time-dependent exposure.⁸

The study by Streja and colleagues addresses a major gap in our understanding of statin use in advanced CKD. The study results are in line with the current recommendation based on expert opinion to continue statin therapy through the initiation of dialysis.⁴ However, observational studies are limited by the available data, and even the most sophisticated statistical methods may not always address some of those limitations. While these data are not sufficient for definitive change to current clinical practice, they offer important supporting evidence. This is likely the best we have until high-quality randomized clinical trials evaluating this critical question are conducted.

Published: October 5, 2018. doi:10.1001/jamanetworkopen.2018.2411

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2018 Edmonston D et al. JAMA Network Open.

Corresponding Author: Michael Pencina, PhD, Department of Biostatistics and Bioinformatics, Duke University School of Medicine, 200 Trent Dr, M144 Davison Bldg, Durham, NC 27710 (michael.pencina@duke.edu).

Conflict of Interest Disclosures: Dr Pencina reported receiving grants from Sanofi/Regeneron, speaking fees from Merck, and consulting fees from Boehringer Ingelheim. No other disclosures were reported.