Low cholesterol in dialysis patients causal factor for mortality or an effect of confounding?

Nephrol Dial Transplant (2011) 26: 3325 3331 doi: 10.1093/ndt/gfr008 Advance Access publication 28 February 2011 Low cholesterol in dialysis patients causal factor for mortality or an effect of confounding? Michal Chmielewski 1, *, Marion Verduijn 2, *, Christiane Drechsler 3, Bengt Lindholm 4, Peter Stenvinkel 5, Boleslaw Rutkowski 1, Elisabeth W. Boeschoten 6, Raymond T. Krediet 7 and Friedo W. Dekker 2 1 Department of Nephrology, Transplantology and Internal Medicine, Medical University of Gdansk, Gdansk, Poland, 2 Department of Clinical Epidemiology, Leiden University Medical Centre, Leiden, The Netherlands, 3 Department of Medicine, Division of Nephrology, University of Würzburg, Würzburg, Germany, 4 Division of Renal Medicine and Baxter Novum, Karolinska Institutet, Stockholm, Sweden, 5 Division of Renal Medicine, Karolinska Institutet, Stockholm, Sweden, 6 Hans Mak Institute, Naarden, The Netherlands and 7 Department of Nephrology, Academic Medical Centre, Amsterdam, The Netherlands Correspondence and offprint requests to: Michal Chmielewski; E-mail: chmiel@gumed.edu.pl *Both authors equally contributed to this work. Abstract Background. The association between cholesterol and mortality is reversed in end-stage renal disease (ESRD). This phenomenon has many potential explanations, one of them being time discrepancy of competing risks. It states that hypercholesterolaemia is beneficial only in the short term, while it worsens survival over a long-term interval. It is also proposed that the reversed relationship between cholesterol and outcome is due to confounding protein energy wasting. The aim of the study was to verify the hypothesis of time discrepancy of competing risks in 1191 incident dialysis patients (Netherlands Cooperative Study on the Adequacy of Dialysis). Methods. Conditional Cox proportional hazards analysis was applied, where associations between cholesterol level and short-term versus long-term mortality were being compared. Furthermore, to evaluate associations between cholesterol and outcome free from confounding, Mendelian randomization was introduced, using apolipoprotein E (apoe) genotype. Results. Hypercholesterolaemia (>240 mg/dl) was associated with improved 5-year survival when compared to the low cholesterol group (<200 mg/dl), hazard ratio (HR) ¼ 0.62 (0.47 0.82), P < 0.001. However, conditional Cox proportional hazards analysis revealed that the reverse association between high cholesterol and all-cause mortality was evident only during the first year of follow-up, HR ¼ 0.43 (0.23 0.80), P < 0.01, and then, gradually, declined. The apoe genotype significantly affected cholesterol concentration. The e2 carriers, associated with low cholesterol, had significantly increased risk of non-cardiovascular mortality. Conclusions. Reverse association between cholesterol concentration and mortality in dialysis patients is short-termed, consistent with the hypothesis of time discrepancy of competing risks. Low cholesterol appeared to affect noncardiovascular mortality in ESRD patients free from confounders. Keywords: cholesterol; dialysis; Mendelian randomization; reverse epidemiology Introduction Patients with end-stage renal disease (ESRD) are at increased risk of mortality, both from cardiovascular and non-cardiovascular causes [1]. There are numerous risk factors that contribute to this condition, including: oxidative stress, calcium phosphate disorders, anaemia, inflammation, protein energy wasting (PEW), lipoprotein disturbances, etc. However, some of these risk factors seem to have an opposite effect on outcome compared to what has been observed in the general population. This reverse association between a risk factor and mortality is especially prominent for cholesterol levels. Several studies have shown that low cholesterol is related to poor outcome in ESRD subjects, while hypercholesterolaemia seems to be protective [2 5]. There are many potential explanations for this, so-called reverse epidemiology phenomenon. Most of them regard inflammation, PEW and other comorbidities as factors responsible for low cholesterol levels and, at the same time, responsible for increased mortality of ESRD patients [4, 6]. Another hypothesis, termed time discrepancy of competing risks, discerns two types of risk factors for increased mortality, short-term and long-term [7]. While hypercholesterolaemia is a significant risk factor for atherosclerosis progression and mortality over a long period of time (decades), as seen in the general population, inflammation and PEW are short-term killers (months years). As the mortality rate for ESRD patients is much higher than for the general population, they do not live long enough for the long-term risk factors to manifest their influence. To add to this confusion, there are also theories of a direct Ó The Author 2011. Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com

3326 M. Chmielewski et al. beneficial impact of increased lipid concentrations on survival in chronic disease states, like ESRD [8]. Whether low cholesterol is a causal risk factor in the ESRD population or is merely a risk marker is difficult to clarify on the basis of observational evaluations. There is, however, a way to substantiate a potential causal role of low cholesterol by the use of a genetic variant related to cholesterol levels in a, so-called Mendelian randomization approach [9]. The gene for apolipoprotein E (apoe), which is polymorphic with three common alleles (e2, e3 and e4), affects cholesterol concentration, with e2 being associated with the lowest and e4 with the highest cholesterol level. Due to random assortment of genes during gamete formation, independent of environmental factors, with the use of apoe variants, the association between cholesterol levels and mortality can be estimated free from confounding and reverse causation. Thus, low cholesterol could be regarded as a causal mortality risk factor if the genetic variant encoding for low cholesterol (e2) is associated with worse outcome. Mendelian randomization approach has been initially proposed by Katan for cancer patients [10] and could similarly be applied in ESRD subjects [9]. The aim of this study was to elucidate the issue of the apparent association between low cholesterol and mortality. The first part of the evaluation implements conditional Cox proportional hazards analysis in order to verify the hypothesis of time discrepancy of competing risks [11]. The second part utilizes associations between genetic variants of apoe and cholesterol levels to get insight into the causative role of low cholesterol using Mendelian randomization. Materials and Methods Patients This study was performed in the Netherlands Cooperative Study on the Adequacy of Dialysis (NECOSAD). This is a multicentre prospective follow-up study in which ESRD patients from 38 Dutch dialysis centres are included at the start of chronic dialysis treatment. All local medical ethics committees approved the study and all patients gave informed consent before inclusion. Eligibility criteria were 18 years and no previous renal replacement therapy. Details on the design and methods of NECO- SAD are described elsewhere [12]. In the present study, patients were included with a minimal follow-up of 3 months since the start of dialysis (baseline of the analysis). From a total group of 1615 patients, we excluded subjects with statin use since, in our opinion, the impact of statins on cholesterol blurred the association between mortality and cholesterol as a marker. This resulted in a cohort of 1269 patients; for 78 of them, no cholesterol measurement was available, so a group of 1191 patients was used for evaluation. The second part of the study, aiming at apoe polymorphism evaluation, included 563 patients, of the 1191 subjects characterized above, who had started dialysis in 1 of the 23 centres that approved DNA analysis. All patients were followed until the date of death or censoring, i.e. kidney transplantation, transfer to a non-participating dialysis centre, withdrawal from the study or end of the follow-up period, with a maximum follow-up of 5 years since start of dialysis. Demographic and clinical data At the start of dialysis, data were collected on age, gender, primary kidney disease, comorbidity and dialysis modality. At 3 months after dialysis initiation, medication use was recorded and subjective global assessment (SGA) was performed. Blood samples were obtained at this time point. Total cholesterol level was measured routinely at the local laboratories, while highsensitive C-reactive protein (hscrp) was measuredcentrally at the University Hospital Aachen, Germany. Clinical definitions Primary kidney disease and causes of death were classified according to the code of the European Renal Association European Dialysis and Transplant Association (ERA EDTA) [13]. Cardiovascular mortality was defined as death due to the following causes: myocardial ischaemia and infarction, hyperkalaemia, hypokalaemia, cardiac arrest, heart failure, fluid overload, cerebrovascular accident, haemorrhage from a ruptured vascular aneurysm, mesenteric infarction and cause of death uncertain/unknown. All other causes of death were designated as non-cardiovascular mortality. The most prevalent were infections, malignancies, cachexia and refusal of further treatment. Patients with a history of angina pectoris, myocardial infarction, cardiac insufficiency or who had peripheral vessel damage or a cerebrovascular accident were classified as having cardiovascular disease. Patients with an hscrp value of >10 mg/l were considered as inflamed. This cutoff point was previously used to divide patients in a group with or without a chronic inflammatory state [14], and in ESRD patients, this cutoff point for C-reactive protein (CRP) level has been shown to be the best with regard to the prediction of survival [15]. Patients with a score of 1 5 on a 7-point SGA were diagnosed as having PEW [16]. ApoE genotyping ApoE genotyping was performed in 434 patients with a DNA sample by polymerase chain reaction (PCR) analysis using primers: ApoE forward 5#-AGA-CGC-GGG-CAC-GGC-TGT-CCA-AGG-A-3# and ApoE reverse 5#-ATA-AAT-ATT-ATA-AAA-TAT-AAA-TAA-CAG-AAT-TCG-CCC- CGG-CCT-GGT-ACA-C-3#. The reaction mixture included 30 pmol of each primer, 100 ng genomic DNA, 0.2 mmol/l of each deoxyribonucleotide triphosphate dntp (Pharmacia), 5 ll 103 PCR buffer (Qiagen), 10% dimethyl sulphoxide, 1.5 U Taq polymerase (Qiagen) in a total volume of 50 ll. Amplification was performed for 35 cycles of 1 min at 95 C, 1 min at 60 C and 2 min at 72 C with an initial denaturation period of 5 min. After the last cycle, a final extension period of 5 min at 72 C min was used. A PCR product of 267 bp was generated. Some 35 ll of PCR products were digested with the restriction enzyme CfoI according to the recommendations of the supplier (Promega). Thereafter, fragments were separated on a 5% agarose gel and stained with ethidium bromide. The e2 allele was characterized by the presence of 104, 91, 38, 18 and 16 bp fragments. The e3allele was characterized by the presence of 91, 56, 48, 38, 18 and 16 bp fragments. The e4 allele was characterized by the presence of 72, 56, 48, 38, 19, 18 and 16 bp fragments. For 129 patients without a DNA sample, apoe phenotypes were determined by isoelectric focusing of delipidated serum samples followed by immunoblotting with polyclonal anti-apoe antiserum as described by Havekes et al. [17]. Statistical analysis All values are expressed as percentages (for categorical variables), mean and SD, unless otherwise indicated. Differences among the groups were analysed by one-way analysis of variance for continuous variables and a chisquare test for categorical variables. For comparisons according to cholesterol level, patients were grouped following the National Cholesterol Education Program - Adult Treatment Panel III guidelines [18] into: low (desirable), with a total cholesterol concentration defined as <200 mg/ dl; borderline, where values varied between 200 and 240 mg/dl and high, with cholesterol >240 mg/dl. We used Kaplan Meier survival analysis to obtain survival curves for each cholesterol category. Hazard ratios (HRs) were calculated by Cox proportional hazards analysis using the low cholesterol group as a reference category. Multivariate adjustment was performed for the following potential confounding factors: age, gender, diabetes mellitus, dialysis modality, inflammation, PEW, haemoglobin, use of erythropoiesis-stimulating agents (ESA), dialysis adequacy (Kt/V) and smoking. We assessed the time-stratified effect of baseline cholesterol on mortality using conditional Cox regression analysis [11] in order to evaluate the influence of duration of follow-up on the association between cholesterol level and mortality. In this analysis, the HRs for mortality during the first year of dialysis were compared to the HRs for mortality during the second year of therapy, conditional upon having survived the first year. Similar evaluation was performed for the Years 3 5, conditional upon having survived the previous period. Therefore, for the long-term survival analysis, we considered only these patients who survived past the first 2 years of dialysis treatment. For the genetic part of the analysis, patients were grouped according to apoe genotypes. The apoe2 group consisted of e2 carriers (e2/2 and e2/3) and the apoe4 of e4 carriers(e3/4 and e4/4), while the reference group (apoe3) consisted of e3/3 homozygotes. Twelve patients with e2/4 genotype

Low cholesterol in dialysis patients 3327 were excluded from the analysis since in these patients, the two alleles affected cholesterol concentration in opposite directions. This resulted in 551 patients for further evaluation. Differences in baseline characteristics among the genotype groups as well as the association between apoe genotype and mortality were evaluated as above. Statistical analyses were performed using statistical software SPSS version 14.0.2 (SPSS Inc., Chicago, IL). Results Basic characteristics of patients according to their cholesterol concentration are presented in Table 1. Their primary kidney disease was renal vascular disease in 17% of cases, diabetic nephropathy in 14% and glomerulonephritis in 13%. For the remaining 56%, ESRD was due to other diseases or unknown causes. The majority of the 1191 patients presented with low (desirable) cholesterol values (55%). Borderline values were observed in 27% of subjects, while 18% of patients displayed high cholesterol concentration according to NCEP ATP III. Patients with low cholesterol levels were older and had a higher prevalence of cardiovascular disease, inflammation and PEW. Male gender and haemodialysis treatment also prevailed in this group. The follow-up time was 5 years with a median observation time of 2.3 years. During this time period, 438 subjects died, 208 of them due to cardiovascular causes. Kaplan Meier analysis showed a considerable association between cholesterol levels and mortality, with low (desirable) cholesterol predicting poor outcome (Figure 1). Indeed, in Cox proportional hazards analysis, hypercholesterolaemia was associated with a significantly decreased risk for all-cause mortality; crude HR ¼ 0.62 (0.47 0.82), P < 0.001. Conditional Cox proportional hazards analysis revealed that the protective association between high cholesterol and survival was evident only during the first year of follow-up; crude HR ¼ 0.43 (0.23 0.80), P < 0.01. Afterward, this relationship was substantially weaker, with HR ¼ 0.67 (0.38 1.17), P ¼ 0.16 in the second year and HR ¼ 0.73 (0.50 1.06), P ¼ 0.12 during the following 3 years (Table 2). After adjustment for potential confounders (age, gender, diabetes mellitus, dialysis modality, inflammation and PEW), the time discrepancy of competing risks became even more evident: hypercholesterolaemia seemed Table 1. Characteristics of the incident dialysis patients included in the study subdivided according to cholesterol concentration a to confer an apparent benefit on outcome only in the first year, HR: 0.47 (0.18 1.23), P ¼ 0.12, while the HRs were much closer to 1 (representing no benefit) in subsequent years (Table 2). Further adjustment for haemoglobin level, use of ESA, dialysis adequacy (Kt/V) and smoking did not alter the results significantly. In further analysis, cardiovascular and non-cardiovascular mortality were evaluated separately as they showed different patterns of association with cholesterol level. For cardiovascular mortality, the association between cholesterol and 5-year outcome did not reach statistical significance; crude HR ¼ 0.75 (0.50 1.10), P ¼ 0.14. For non-cardiovascular mortality, it was considerably stronger; crude HR ¼ 0.53 (0.35 0.78), P < 0.01, although substantially attenuated following adjustment for age, gender, diabetes mellitus, dialysis modality, inflammation and PEW; HR ¼ 0.76 (0.44 1.32), P ¼ 0.34. In conditional Cox analysis, cardiovascular mortality was significantly affected by cholesterol concentration only during the first year of dialysis therapy (Table 3). During the following years of observation, this association disappeared. For non-cardiovascular mortality, benefit of hypercholesterolaemia was observed across the whole follow-up period. The association between cholesterol and noncardiovascular mortality was, however, deeply affected by adjustment for confounders. From the second year onwards, the relationship vanished following adjustment for comorbidities, inflammation and PEW (Table 4). It is worth mentioning that there was only one death due to infectious causes in the high cholesterol group, during the first 2 years of dialysis treatment. General characteristics of the patients grouped according to their apoe genotype are presented in Table 5. The e2 carriers were slightly older than the other two groups of patients, while e4 carriers were less inflamed and less wasted than e2 subjects. Similarly to the general population, cholesterol concentration was significantly lower in patients with the e2 variant. Since low cholesterol level was associated with poor outcome in the first part of the analysis, we tested whether the e2 variant would be related to increased mortality. The median follow-up time was 2.4 years; during this time period, 199 patients died, 91 due to cardiovascular Cholesterol concentration (mg/dl) <200 Low b (N ¼ 657) 200 240 Borderline b (N ¼ 325) >240 High b (N ¼ 209) P-value Age, years 60.4 (16.4) 58.4 (15.7) 57.3 (14.4) <0.01 Gender, % male 67 58 49 <0.01 CVD, % 36 26 27 <0.01 DM, % 20 18 19 0.76 Inflamed, % (CRP > 10 mg/l) 33 29 20 0.02 Serum albumin (g/l) 36.0 (32.2 39.0) 37.1 (34.0 40.0) 37.4 (34.0 41.0) <0.01 Wasted, % (SGA 5) 35 24 30 <0.01 Dialysis modality, % HD 75 56 43 <0.01 a CVD, cardiovascular disease; DM, diabetes mellitus, HD, haemodialysis. b According to NCEP ATP III [18].

3328 M. Chmielewski et al. causes. Although no difference in all-cause mortality risk was observed among the apoe groups, the e2 variant was associated with substantially increased risk of noncardiovascular mortality; HR ¼ 1.61 (1.02 2.55), P ¼ 0.04, while the risk of cardiovascular mortality was decreased in this patient group; HR: 0.70 (0.38 1.30), P ¼ 0.26 (Figure 2). Adjustment for age resulted in a slight decrease of the risk for both non-cardiovascular and cardiovascular mortality, with HRs being 1.38 and 0.60, respectively. Discussion Fig. 1. Five-year patient survival in each of cholesterol concentration groups. Table 2. Conditional Cox proportional hazards analysis for all-cause mortality a All patients (N ¼ 1191) In the present study, we have shown that the reverse association between cholesterol concentration and mortality in dialysis patients is short-termed, which is consistent with Cholesterol category Low (N ¼ 657) Borderline (N ¼ 325) High (N ¼ 209) First year (121 events) Crude HR 1 0.78 (0.52 1.18); P ¼ 0.24 0.43 (0.23 0.80); P < 0.01 Adjusted 1 1 0.94 (0.62 1.44); P ¼ 0.77 0.59 (0.31 1.12); P ¼ 0.11 Adjusted 2 1 0.96 (0.53 1.73); P ¼ 0.88 0.47 (0.18 1.23); P ¼ 0.12 Adjusted 3 1 0.97 (0.53 1.77); P ¼ 0.91 0.34 (0.11 0.99); P ¼ 0.05 Second year (106 events) Crude HR 1 0.85 (0.54 1.33); P ¼ 0.48 0.67 (0.38 1.17); P ¼ 0.16 Adjusted 1 1 0.95 (0.60 1.51); P ¼ 0.83 0.83 (0.46 1.50); P ¼ 0.54 Adjusted 2 1 1.16 (0.64 2.11); P ¼ 0.62 1.09 (0.52 2.29); P ¼ 0.83 Adjusted 3 1 1.15 (0.61 2.16); P ¼ 0.67 1.14 (0.52 2.52); P ¼ 0.74 Following 3 years (212 events) Crude HR 1 0.77 (0.55 1.06); P ¼ 0.11 0.73 (0.50 1.06); P ¼ 0.12 Adjusted 1 1 0.79 (0.56 1.11); P ¼ 0.17 0.88 (0.59 1.30); P ¼ 0.51 Adjusted 2 1 0.70 (0.44 1.11); P ¼ 0.13 0.93 (0.54 1.59); P ¼ 0.78 Adjusted 3 1 0.65 (0.40 1.07); P ¼ 0.09 0.95 (0.54 1.67); P ¼ 0.95 a Adjusted 1 for age, gender, diabetes mellitus and dialysis modality; adjusted 2 above 1 inflammation (hscrp > 10 mg/l) and PEW (SGA 1 5) and adjusted 3 above 1 dialysis adequacy (Kt/V), use of ESA, haemoglobin and smoking. Table 3. Conditional Cox proportional hazards analysis for cardiovascular mortality a All patients (N ¼ 1191) Cholesterol category Low (N ¼ 657) Borderline (N ¼ 325) High (N ¼ 209) First year (51 events) Crude HR 1 0.93 (0.51 1.72); P ¼ 0.82 0.38 (0.14 1.08); P ¼ 0.07 Adjusted 1 1 1.06 (0.57 1.99); P ¼ 0.87 0.47 (0.16 1.37); P ¼ 0.17 Adjusted 2 1 1.50 (0.59 3.81); P ¼ 0.40 0.60 (0.13 2.85); P ¼ 0.52 Adjusted 3 1 1.55 (0.61 3.97); P ¼ 0.36 0.22 (0.03 1.87); P ¼ 0.17 Second year (51 events) Crude HR 1 1.32 (0.71 2.43); P ¼ 0.38 0.88 (0.40 1.94); P ¼ 0.85 Adjusted 1 1 1.53 (0.82 2.85); P ¼ 0.19 1.16 (0.51 2.67); P ¼ 0.72 Adjusted 2 1 1.67 (0.74 3.76); P ¼ 0.22 1.09 (0.35 3.44); P ¼ 0.88 Adjusted 3 1 1.61 (0.67 3.89); P ¼ 0.29 1.40 (0.43 4.06); P ¼ 0.58 Following 3 years (107 events) Crude HR 1 0.84 (0.53 1.34); P ¼ 0.47 0.90 (0.55 1.48); P ¼ 0.68 Adjusted 1 1 0.88 (0.55 1.42); P ¼ 0.60 1.09 (0.64 1.84); P ¼ 0.76 Adjusted 2 1 0.51 (0.25 1.03); P ¼ 0.06 0.98 (0.47 2.02); P ¼ 0.98 Adjusted 3 1 0.48 (0.23 0.99); P ¼ 0.05 1.00 (0.48 2.01); P ¼ 0.99 a Adjusted 1 for age, gender, diabetes mellitus and dialysis modality; adjusted 2 above 1 inflammation (hscrp > 10 mg/l) and PEW (SGA 1 5) and adjusted 3 above 1 dialysis adequacy (Kt/V), use of ESA, haemodialysis and smoking.

Low cholesterol in dialysis patients 3329 Table 4. Conditional Cox proportional hazards analysis for non-cardiovascular mortality a All patients (N ¼ 1191) Cholesterol category Low (N ¼ 657) Borderline (N ¼ 325) High (N ¼ 209) First year (70 events) Crude HR 1 0.68 (0.38 1.19); P ¼ 0.18 0.45 (0.21 1.00); P ¼ 0.05 Adjusted 1 1 0.85 (0.48 1.52); P ¼ 0.59 0.68 (0.30 1.54); P ¼ 0.36 Adjusted 2 1 0.71 (0.32 1.57); P ¼ 0.40 0.41 (0.12 1.41); P ¼ 0.16 Adjusted 3 1 0.70 (0.31 1.58); P ¼ 0.40 0.40 (0.11 1.40); P ¼ 0.15 Second year (55 events) Crude HR 1 0.53 (0.27 1.07); P ¼ 0.08 0.52 (0.23 1.17); P ¼ 0.12 Adjusted 1 1 0.57 (0.28 1.16); P ¼ 0.12 0.60 (0.26 1.42); P ¼ 0.25 Adjusted 2 1 0.79 (0.32 1.96); P ¼ 0.62 1.08 (0.40 2.90); P ¼ 0.88 Adjusted 3 1 0.81 (0.32 2.07); P ¼ 0.66 0.96 (0.33 2.82); P ¼ 0.95 Following 3 years (105 events) Crude HR 1 0.70 (0.44 1.11); P ¼ 0.13 0.58 (0.33 1.01); P ¼ 0.05 Adjusted 1 1 0.71 (0.44 1.14); P ¼ 0.16 0.69 (0.38 1.24); P ¼ 0.22 Adjusted 2 1 0.93 (0.49 1.75); P ¼ 0.82 0.85 (0.38 1.93); P ¼ 0.70 Adjusted 3 1 0.91 (0.46 1.77); P ¼ 0.77 0.88 (0.37 2.11); P ¼ 0.78 a Adjusted 1 for age, gender, diabetes mellitus and dialysis modality; Adjusted 2 above 1 inflammation (hscrp > 10 mg/l) and PEW (SGA 1 5) and Adjusted 3 above 1 dialysis adequacy (Kt/V), use of ESA, haemoglobin and smoking. Table 5. Characteristics of the incident dialysis patients included in the study subdivided according to the apoe gene isoform a ApoE variant e2 (N ¼ 100) e3 (N ¼ 322) e4 (N ¼ 129) P-value Cholesterol, mg/dl 189.0 (45.7) 195.5 (48.8) 208.4 (50.7) <0.01 Age, years 62.1 (15.6) 59.4 (16.1) 56.6 (15.3) 0.04 Gender, % male 59 62 63 0.84 CVD, % 30 29 34 0.65 DM, % 21 19 16 0.62 Inflamed, % (CRP > 10 mg/l) 35 33 23 0.16 Wasted, % (SGA 5) 40 32 25 0.06 Dialysis modality, % HD 69 64 55 0.08 a CVD, cardiovascular disease; DM, diabetes mellitus, HD, haemodialysis. Fig. 2. HRs for the association between the apoe genotype and 5 years mortality in 551 patients. the hypothesis of time discrepancy of competing risks. Moreover, in the studied cohort, subjects with genetically determined lower cholesterol concentrations were at increased risk of non-cardiovascular mortality. Before discussing these results, some caveats of the study should be mentioned. First, the sample size was rather small, especially for the genetic part of the study. This was associated with the relatively low number of events, which resulted in wide confidence intervals in survival analyses. Nevertheless, NECOSAD is one of the largest European cohorts of incident dialysis patients. Another potential limitation might be that for predicting outcome in the general population, total cholesterol is inferior to low-density lipoprotein cholesterol or apob/apoa-i. Unfortunately, these lipid parameters, as well as high-density lipoprotein cholesterol, were not collected in the NECOSAD cohort. The most persuasive explanation for the observed reverse association between low cholesterol levels and mortality in ESRD is that the phenomenon is due to confounding by inflammation and/or PEW. This hypothesis was supported by Liu et al. [4] who showed that low cholesterol predicted mortality only in patients with low serum albumin levels (i.e. inflamed and/or wasted). The subgroup of patients with normal albumin concentrations displayed a typical association between high cholesterol and poor outcome. However, a subsequent study [5] in a similar cohort of patients did not confirm the results of Liu

3330 M. Chmielewski et al. et al. [4]. Moreover, large randomized clinical trials in ESRD patients showed no benefit of cholesterol reduction despite clear effects of statin therapy on outcome in the general population [19, 20]. Since the issue remained opened, theories on a direct beneficial impact of increased cholesterol concentration started to rise. A theory put forward by Rauchhaus et al. [21] stated that in certain clinical conditions low cholesterol might actually be harmful. This hypothesis has been supported by results of experimental studies showing that lipoproteins are capable of binding bacterial lipopolysaccharide or endotoxins [22 24]. Thus, the endotoxin lipoprotein theory might be applicable in dialysis patients, who are at increased risk of infections because of immunological disturbances and the dialysis procedure itself. One of the most convincing theories explaining the phenomenon of reverse epidemiology is the time discrepancy of competing risks hypothesis. It has been proposed by Kalantar-Zadeh et al. [7] and states that hypercholesterolaemia appears beneficial only in the short-term, while over a long follow-up period, it becomes detrimental. As mentioned, it is based on the fact that life expectancy for the general population is much longer than for dialysis patients. Whereas subjects from the general population live long enough for high cholesterol levels to have an impact on atherosclerosis and cardiovascular complications, dialysis patients have a life expectancy comparable to patients with colon cancer. Therefore, in the short-term, ESRD patients die because of disorders associated with inflammation and/ or PEW, and we can observe an association between low cholesterol and mortality. We have previously verified this hypothesis in a cohort of Swedish incident dialysis patients [25] showing that whereas increased apob/apoa-1 ratio predicted short-term (first year) survival, it predicted longterm (next 3 years) mortality. In order to verify whether observed associations are causal or simply due to confounders, we have introduced adjustments to survival analyses. Indeed, adding consecutive potential confounders into Cox analysis resulted in an attenuation of presented relationships. However, especially in the first year of follow-up, even extensive adjustment did not level the associations completely. It might be due to residual confounders, the one that we did not account for, or to potential direct effect of cholesterol on outcome. To clarify this issue, we also analysed our data using the Mendelian randomization approach with the apoe genotype as an instrumental variable. The use of this method to draw causal inferences from observational studies is becoming increasingly common [26]. Our analysis showed that e2 carriers (low cholesterol) had a slightly decreased risk of cardiovascular mortality, which seems logical since low cholesterol is typically associated with slower atherosclerosis progression. However, as the risk of noncardiovascular mortality was increased in e2 carriers, low cholesterol might directly increase the risk of noncardiovascular complications in ESRD patients. This supports the endotoxin lipoprotein hypothesis stating that decreased lipoprotein concentrations are associated with impaired defense against bacterial endotoxins. Indeed, only one patient died of infectious causes in the high cholesterol group during the first 2 years of dialysis therapy. Conclusions In conclusion, reversed association between cholesterol concentration and mortality is mainly short-termed, which is consistent with the hypothesis of time discrepancy of competing risks. It is, to a significant extent, a result of confounding factors such as age, comorbidities, inflammation and PEW. Our results also suggest the existence of a causal relationship between low cholesterol and non-cardiovascular mortality. Acknowledgements. This study was supported by grants from the Dutch Kidney Foundation (E.018) and the Dutch National Health Insurance Board (OG97/005). The nursing staffs of the participating dialysis centres are gratefully acknowledged for collecting most of the clinical data. The authors also wish to thank the staff of the NECOSAD trial office for their assistance in the logistics of this study. Conflict of interest statement. None declared. References 1. de Jager DJ, Grootendorst DC, Jager KJ et al. Cardiovascular and noncardiovascular mortality among patients starting dialysis. JAMA 2009; 302: 1782 1789 2. Lowrie EG, Lew NL. Death risk in hemodialysis patients: the predictive value of commonly measured variables and an evaluation of death rate differences between facilities. Am J Kidney Dis 1990; 15: 458 482 3. Iseki K, Yamazato M, Tozawa M et al. Hypocholesterolemia is a significant predictor of death in a cohort of chronic hemodialysis patients. Kidney Int 2002; 61: 1887 1893 4. Liu Y, Coresh J, Eustace JA et al. Association between cholesterol level and mortality in dialysis patients: role of inflammation and malnutrition. JAMA 2004; 291: 451 459 5. Kilpatrick RD, McAllister CJ, Kovesdy CP et al. Association between serum lipids and survival in hemodialysis patients and impact of race. J Am Soc Nephrol 2007; 18: 293 303 6. Fleischmann EH, Bower JD, Salahudeen AK. Risk factor paradox in hemodialysis: better nutrition as a partial explanation. ASAIO J 2001; 47: 74 81 7. Kalantar-Zadeh K, Block G, Humphreys MH et al. Reverse epidemiology of cardiovascular risk factors in maintenance dialysis patients. Kidney Int 2003; 63: 793 808 8. Chmielewski M, Carrero JJ, Nordfors L et al. Lipid disorders in chronic kidney disease: reverse epidemiology and therapeutic approach. J Nephrol 2008; 21: 635 644 9. Zoccali C, Testa A, Spoto B et al. Mendelian randomization: a new approach to studying epidemiology in ESRD. Am J Kidney Dis 2006; 47: 332 341 10. Katan MB. Apolipoprotein E isoforms, serum cholesterol, and cancer. Lancet 1986; 1: 507 508 11. Dekker FW, de Mutsert R, van Dijk PC et al. Survival analysis: timedependent effects and time-varying risk factors. Kidney Int 2008; 74: 994 997 12. Muntinghe FL, Verduijn M, Zuurman MW et al. CCR5 deletion protects against inflammation-associated mortality in dialysis patients. J Am Soc Nephrol 2009; 20: 1641 1649 13. van Dijk PC, Jager KJ, de Charro F et al. Renal replacement therapy in Europe: the results of a collaborative effort by the ERA-EDTA registry and six national or regional registries. Nephrol Dial Transplant 2001; 16: 1120 1129 14. Grootendorst DC, de Jager DJ, Brandenburg VM et al. Excellent agreement between C-reactive protein measurement methods in end-stage renal disease patients no additional power for mortality prediction with high-sensitivity CRP. Nephrol Dial Transplant 2007; 22: 3277 3284 15. Stenvinkel P, Wanner C, Metzger T et al. Inflammation and outcome in end-stage renal failure: does female gender constitute a survival advantage? Kidney Int 2002; 62: 1791 1798

Low cholesterol in dialysis patients 3331 16. Steiber A, Leon JB, Secker D et al. Multicenter study of the validity and reliability of subjective global assessment in the hemodialysis population. J Ren Nutr 2007; 17: 336 342 17. Havekes LM, de Knijff P, Beisiegel U et al. A rapid micromethod for apolipoprotein E phenotyping directly in serum. J Lipid Res 1987; 28: 455 463 18. Third Report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) final report. Circulation 2002; 106: 3143 3421 19. Wanner C, Krane V, Marz W et al. Atorvastatin in patients with type 2 diabetes mellitus undergoing hemodialysis. N Engl J Med 2005; 353: 238 248 20. Fellstrom BC, Jardine AG, Schmieder RE et al. Rosuvastatin and cardiovascular events in patients undergoing hemodialysis. N Engl J Med 2009; 360: 1395 1407 21. Rauchhaus M, Coats AJ, Anker SD. The endotoxin-lipoprotein hypothesis. Lancet 2000; 356: 930 933 22. Flegel WA, Baumstark MW, Weinstock C et al. Prevention of endotoxin-induced monokine release by human low- and high-density lipoproteins and by apolipoprotein A-I. Infect Immun 1993; 61: 5140 5146 23. Levine DM, Parker TS, Donnelly TM et al. In vivo protection against endotoxin by plasma high density lipoprotein. Proc Natl Acad Sci U S A 1993; 90: 12040 12044 24. Read TE, Grunfeld C, Kumwenda ZL et al. Triglyceride-rich lipoproteins prevent septic death in rats. JExpMed1995; 182: 267 272 25. Chmielewski M, Carrero JJ, Qureshi AR et al. Temporal discrepancies in the association between the apob/apoa-i ratio and mortality in incident dialysis patients. J Intern Med 2009; 265: 708 716 26. Davey Smith G, Ebrahim S. Mendelian randomization : can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol 2003; 32: 1 22 Received for publication: 27.9.10; Accepted in revised form: 5.1.11