Dyslipidemia and risk of renal replacement therapy or death in incident pre-dialysis patients

Globally the number of patients on renal replacement therapy (RRT) is rising. Dyslipidemia is a potential modifiable cardiovascular risk factor, but its effect on risk of RRT or death in pre-dialysis patients is unclear. The aim of this study was to assess the association between dyslipidemia and risk of RRT or death among patients with CKD stage 4–5 receiving specialized pre-dialysis care, an often under represented group in clinical trials. Of the 502 incident pre-dialysis patients (>18 y) in the Dutch PREPARE-2 study, lipid levels were available in 284 patients and imputed for the other patients. During follow up 376 (75%) patients started RRT and 47 (9%) patients died. Dyslipidemia was defined as total cholesterol ≥5.00 mmol/L, LDL cholesterol ≥2.50 mmol/L, HDL cholesterol <1.00 mmol/L, HDL/LDL ratio <0.4, or triglycerides (TG) ≥2.25 mmol/L, and was present in 181 patients and absent in 93 patients. After multivariable adjustment Cox regression analyses showed a HR (95% CI) for the combined endpoint for dyslipidemia of 1.12 (0.85–1.47), and for high LDL of 1.20 (0.89–1.61). All other HRs were smaller. In conclusion, we did not find an association between dyslipidemia or the separate lipid levels and RRT or death in CKD patients on specialized pre-dialysis care.


Sensitivity analyses.
Adding kidney function to model 3 changed the HR for start of dialysis for dyslipi- for the HDL/LDL ratio, and to 0.68 (0.44-1.04) for triglycerides. The HRs for the combined endpoint for total cholesterol and HDL did not change when analyzed as a categorical variable. The HRs for the combined endpoint when analyzing lipids as continuous variables did not change. When we confined all analyses to patients with at least one cholesterol measurement during the first 6 months of their study participation the results did not change essentially.
To account for changes of serum lipids over time, we examined the short term (≤12 months) and long term (>12 months) association between serum lipids and outcomes separately. Tables 5 and 6 show the adjusted hazard ratios for the outcomes for dyslipidemia, and for the serum lipids as a categorical and continuous variable for short and long term follow-up. For dyslipidemia, the short term HRs were lower as compared with the long term HRs, 0.97 (0.63 to 1.49) vs 1.23 (0.80 to 1.88) for the combined outcome. For the separate lipid categories and levels short term effects were stronger than long term effects, but without any significant differences since all 95% confidence intervals overlapped.

Discussion
We found no association between dyslipidemia and start of dialysis, RRT or death in incident pre-dialysis patients.
Our finding is in line with several other studies in CKD patients 7,13,17 . The CRIC study (mean age 58 y, mean eGFR 45 ml/min/1.73 m², 42% non-Hispanic black) showed no association between lipoprotein levels and risk of end-stage renal disease (ESRD) after multivariable adjustment 7,17 . Chawla et al. found no association between lipid levels and 10-year mortality in 840 CKD stage 3-4 non diabetic patients (mean age 52 y, mean GFR 33 ml/ min/1.73 m², 85% white patients) 13 .
Kovesdy et al. found in 986 patients (mean age 67 y, mean eGFR 37 ml/min/1.73 m², 40% statin usage) after multivariable adjustment, including controlling for case-mix and malnutrition-inflammation factors, no association between total cholesterol, LDL and TG, and mortality 12 . In our study additional adjustment for malnutrition-inflammation factors did not essentially change the results, which can be explained by the characteristics of our population. Almost all patients of our cohort were well nourished (SGA of 6 or 7), only less than 10% had a moderate nourishment status. Unfortunately, in 25% of all patients the SGA measurement was not performed. Therefore, we cannot exclude that some pre-dialysis patients were severely malnourished. On the other hand, only 2% of all patients in our cohort had a very low BMI <18.5 kg/m 2 . To adjust for residual confounding due to under nutrition we added serum albumin as a nutritional marker to our analyses, which essentially did not change the results. Our results are in concordance with a recent guideline, stating that CKD patients ≥50 y should be treated with a statin, independent of lipid or triglyceride levels, without aiming at a target level 18,19 . This shows a paradigm shift moving away from LDL based therapy, towards treatment based on atherosclerotic cardiovascular risks. This shift is caused by the lack of evidence that changing lipid levels affects cardiovascular risk in CKD patients [20][21][22] .
The guideline advice to start statin therapy independent of lipid levels is based on results of subanalyses of large trials 18,19 . For example the SHARP trial, including 9270 patients with a mean eGFR of 27 ml/min/1.73 m² (mean age 62 y, 63% men, and 23% diabetics), showed that patients treated with statins compared to no statins had a 17% lower risk of cardiovascular outcome 15 25 . Several hypotheses have been suggested to explain the lack of association between dyslipidemia and cardiovascular morbidity and mortality in patients with impaired kidney function. First, uremia may transform HDL into a promotor of inflammation and atherogenesis [26][27][28][29] . In addition, Bauer et al. found that HDL functionality (HDL cholesterol efflux capacity) is not associated with cardiovascular events in CKD patients (mean eGFR 46 ml/min/1.73 m²) 30 . This is in line with our finding that high HDL level was weakly associated with an increased risk of adverse outcome. If uremia induced, one would expect a dose-response relation, and as a result a decreasing beneficial effect of high levels of HDL in the later CKD stages. After stratification for baseline eGFR (≤15 or >15 ml/min/1.73 m²) we found a HR for the combined endpoint of 1.10 (95% CI 0.72 to 1.69) and 0.92 (95% CI 0.54 to 1.58) for high versus low HDL 30 . Second, although we adjusted for malnutrition-inflammation factors, we cannot exclude residual confounding by coexistent wasting and/or inflammation that led to both lower lipid levels and an increased risk of adverse outcome. Finally, severe atherosclerosis, as present in most pre-dialysis patients, is multifactorial, and might be too advanced to achieve disease regression by lower lipid levels. In addition, non-traditional cardiovascular risk factors such as a disturbed calcium-phosphate balance may play a more important role than dyslipidemia in the progression of atherosclerosis [31][32][33] .
Main strength of this study is the specific selection of pre-dialysis patients who were treated according to the current CKD guidelines by nephrologists. Pre-dialysis patients form a special group in chronic kidney disease care and cannot be compared to patients in the early stages of CKD. Since no exclusion criteria were used for the PREPARE cohort a wide range of incident pre-dialysis patients were included and all patient information was used to perform the analyses, our results can be generalized to the clinical practice of pre-dialysis care.
This study has limitations. The main limitation of this study are the missing data. Even though we used multiple imputation to deal with this in the best possible way, it is possible that the amount of predictors was insufficient to complete the data. However, in a sensitivity analysis where multiple imputation was restricted to patients with at least one lipid measurement during the first 6 months of study participation, results did not change materially. Second, information regarding the fasting state of patients was not available, resulting in a lack of distinction between patients with fasting or post-prandial elevated TG levels. Finally we used two different methods to measure LDL. Under normal circumstances the Friedewald equation correlates very well with the direct measurement of LDL 34 . However, we cannot exclude the occurrence of chylomicrons or a combination with high plasma TG, both causes of under-and overestimation of LDL levels.
In conclusion, we found no clear association between dyslipidemia and start of dialysis, RRT or death.

Methods
Study design and population. The PRE-dialysis Patient Record-2 (PREPARE-2) study is a prospective cohort study of incident pre-dialysis care patients (≥18 y) who had an estimated glomerular filtration rate (eGFR) of less than 20-30 ml/min/1.73 m² and progressive renal function loss. Patients with a failing kidney transplant, who were transplanted at least one year ago, were also eligible for inclusion. The study has been described in detail elsewhere 35 . In brief, patients were recruited in one of 25 nephrology specialized pre-dialysis outpatient clinics in the Netherlands between July 2004 and June 2011. All patients were treated by their nephrologist in accordance with the treatment guidelines of the Dutch Federation of Nephrology, guidelines partly based on the K/DOQI and EBPG guidelines [36][37][38][39] . Patients were followed from the start of pre-dialysis care until start of dialysis, kidney transplantation, death or censoring. Censoring was defined as: refusal for further participation, recovery of kidney function, moving to an outpatient clinic not participating in the PREPARE-2 study, loss to follow up or October 2016 (end of follow up), whichever came first. This study was approved by the medical ethics committee or institutional review boards (as appropriate) of all participating centers. Written informed consent was obtained from all patients. All methods were performed in accordance with the relevant guidelines and regulations.
Demographic and clinical data. Data on demography, primary kidney disease, comorbidities, medication use, and laboratory values were collected during routine visits to pre-dialysis outpatient clinics. These visits took place at the start of specialized pre-dialysis care, at the moment of reaching one of the study endpoints as described previously, and every intermediate 6-month interval. Laboratory data were extracted from the electronic hospital information systems or medical records. The closest laboratory measurement performed within 90 days before or after the date of a visit was appointed to that visit. HDL cholesterol and TG levels were directly measured following standard procedure in the participating outpatient clinics. LDL cholesterol was either directly measured or estimated with the    38,45,46 . Baseline characteristics were presented for the total population and according to presence or absence of dyslipidemia. Absolute crude incidence rates of the primary outcomes were calculated for the total population and separately for patients with and without dyslipidemia. We conducted Cox proportional hazards regression analysis, obtaining hazard ratios (HR) with 95% confidence intervals (95% CI) to estimate the effect of dyslipidemia and the different components of dyslipidemia on the three primary outcomes. Because dyslipidemia shows its detrimental effects after long term exposure, we studied dyslipidemia as a fixed risk factor at baseline. The separate components of dyslipidemia were analyzed as categorical and continuous variables. Analyses were adjusted for age, sex, ethnicity, body mass index, diabetes mellitus, hypertension, primary kidney disease, proteinuria and current smoking (model 1). In addition to model 1 we also adjusted for malnutrition-inflammation factors: serum albumin, serum C-Reactive Protein, the SGA score (model 2), as well as for lipid-lowering medication use (statin use, fibrate use, or cholesterol absorption medication use) (model 3). Follow-up time was defined as time between baseline visit of the patient and the start of dialysis, RRT, death, withdrawal or end of follow-up (October 2016). The proportional hazard assumption was tested using a log minus log plot. To estimate the median follow up time, a reversed Kaplan-Meier was used.
Multiple imputation was used to avoid bias and to maintain power 47,48 . Missing values of total cholesterol, LDL cholesterol, HDL cholesterol and TG at baseline, as well as potential confounders at baseline were imputed (using 10 repetitions). The imputed data were predicted based on the available information of each patient.
We performed multiple sensitivity analyses to test the robustness of our findings. First, we added kidney function at baseline into the multivariable models. Since kidney function could be in the causal pathway between Scientific REPORTS | (2018) 8:3130 | DOI:10.1038/s41598-018-20907-y dyslipidemia and the outcomes we did not add this variable in the main model. Second, we stratified for statin use, because statins may have a pleiotropioc, non-lipid lowering effect, independent of the effect on lipid levels. Third, we stratified for baseline eGFR (≤15 vs >15 ml/min/1.73 m²) to study effect modification between kidney functon and dyslipidemia with regard to the outcome. Fourth, we restricted our analysis to patients who were persistent users or non-users of lipid-lowering medication during the entire study period (adjusted for model 3), since changes in lipid-lowering therapy during the follow up period might dilute treatment effects. Fifth, we studied short and long term effects from baseline dyslipidemia seperately by restricting our follow up time to 12 months (short term) and by restricting our analyses to patients who were still in the study after 12 months (long term). Finally, we repeated all analyses applying multiple imputation confined to patients with at least one serum total cholesterol, LDL cholesterol, HDL cholesterol or TG measurement during the first 6 months of their study participation. A p-value < 0.05 was considered statistically significant. All analyses were performed using SPSS version 23.0 for Windows. Data availability. All data generated or analysed during this study are included in this published article. The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.