Article Text

Download PDFPDF

Original research article
Coronary heart disease risk associated with the dyslipidaemia of chronic kidney disease
  1. Julio A Lamprea-Montealegre1,
  2. Robyn L McClelland2,
  3. Morgan Grams3,
  4. Pamela Ouyang4,
  5. Moyses Szklo5,
  6. Ian H de Boer6
  1. 1 Cardiology Division, Department of Medicine, University of Washington School of Medicine, Seattle, Washington, USA
  2. 2 Biostatistics Department, University of Washington School of Public Health, Seattle, Washington, USA
  3. 3 Nephrology Division, Department of Medicine, Johns Hopkins University, Baltimore, Maryland, USA
  4. 4 Cardiology Division, Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA
  5. 5 Epidemiology Department, Johns Hopkins University School of Public Health, Baltimore, Maryland, USA
  6. 6 Nephrology Division, Department of Medicine, University of Washington School of Medicine, Seattle, Washington, USA
  1. Correspondence to Dr Julio A Lamprea-Montealegre, Cardiology Department, University of Washington School of Medicine, Seattle, WA 98195, USA; jlamprea{at}


Objective This study sought to characterise the main dyslipidaemic phenotypes present in chronic kidney disease (CKD) and their association with coronary heart disease (CHD) risk.

Methods Analyses included 6612 individuals in the multiethnic study of atherosclerosis free of CHD at baseline. CKD was defined as an estimated glomerular filtration rate (eGFR) of 15 to <60 mL/min/1.73 m2 (stages 3–4). Principal component analyses were used to characterise the main dyslipidaemic phenotypes of CKD accounting for the correlation among different lipoproteins and lipoprotein particles. CHD was defined as incident myocardial infarction, angina followed by revascularisation, resuscitated cardiac arrest or CHD death.

Results CHD developed in 303 individuals (5%) with eGFR ≥60 and in 72 individuals (12%) with CKD (p for difference <0.001). A dyslipidaemic phenotype (principal component 1 (PC1)) consisting of elevations in triglycerides, triglyceride-rich lipoproteins (VLDL particles), small LDL particles and reductions in HDL particles, was more common in those with CKD, compared with those without CKD (p for difference <0.001). This phenotype was also more strongly associated with CHD in those with CKD: adjusted HRs (95% CIs) per SD increase in PC1 1.13 (95% CI 1.00 to 1.27; P=0.05) and 1.51 (95% CI 1.17 to 1.94; P<0.001) in eGFR ≥60 and CKD, respectively (P for interaction=0.05).

Conclusion In individuals with mainly stage 3 CKD, a dominant lipid phenotype consisting of triglyceride-rich lipoproteins and other closely correlated lipoproteins is strongly associated with CHD risk. Future studies should investigate whether modification of the components of this phenotype leads to a reduction in the CHD burden in individuals with CKD.

  • lipoproteins and hyperlipidemia
  • coronary artery disease
  • epidemiology

Statistics from


Chronic kidney disease (CKD) is a strong, graded and independent risk factor for the development of cardiovascular disease (CVD).1 Most individuals with a CKD diagnosis will develop a cardiovascular outcome before progressing to end stage renal disease (ESRD).2 Despite the strong association between CKD and CVD, knowledge of the mechanisms contributing to this association is lacking.

There has been a recent focus on alterations of triglyceride (TG)-rich lipoproteins as significant cardiovascular disease predictors.3 Mendelian randomisation analyses support the notion of a causal association for TGs and their closely correlated lipoprotein abnormalities including smaller and denser low-density lipoprotein (LDL) particles and high-density lipoprotein (HDL) particles with CVD.3 4 These studies have led to the development of novel non-statin therapies aimed at decreasing the concentration of TGS and their correlates.5–7

In prior studies, we demonstrated that in individuals with mild and moderate reductions in estimated glomerular filtration rate, a dyslipidaemic phenotype consisting of increased levels of TG-rich lipoproteins is highly prevalent and also strongly associated with a high subclinical atherosclerotic burden.8 9 Thus, our study sought to assess the association of lipids and lipoprotein particles with the risk of clinical coronary heart disease (CHD) in individuals with CKD.

Materials and methods

Study population

The multiethnic study of atherosclerosis (MESA) is a multicenter, multiracial, population-based cohort study aimed at studying risk factors for the incidence and progression of cardiovascular disease. Detailed description of its methods have been previously published.10 For the current study, 6784 individuals free of known cardiovascular disease at baseline (visit 1) from 2000 through 2002 were included. Individuals with missing baseline cystatin C measurements (n=58), missing event-time data (n=4), missing measurements for any of the lipid components analysed (n=105) and individuals in whom estimated glomerular filtration rate (eGFR) was <15 mL/min/1.73 m2 (n=5) were excluded. After applying these exclusion criteria, a total of 6612 individuals (97% of the initial cohort) formed the study population. These individuals were followed for a median of 8.5 years for the development of incident CHD.

All participants gave written informed consent prior to the start of the study.10

Kidney function

A cystatin C-based formula for GFR estimation (CKD-EPI cystatin C)11 was used in the primary analyses:

133×min (Scys/0.8, 1)−0.499×max (Scys/0.8, 1)−1.328×0.996Age(×0.932 if female), where Scys is serum cystatin C, min indicates the minimum of Scys/k or 1 and max indicates the maximum of Scys/k or 1.

This formula has been shown to be a precise and accurate method for GFR estimation. It was chosen over creatinine because cystatin C has been shown to be superior to creatinine in predicting cardiovascular events and less subjected to the effects of age, gender and race than creatinine.12 CKD was defined as an eGFR <60 mL/min/1.73 m2.

Lipids and lipoprotein particles

Major lipids and lipoproteins including fasting total cholesterol (TC), HDL cholesterol (HDLc) and TG were measured in all participants using standardised enzymatic methods.13 LDL cholesterol concentrations (LDLc) were estimated using the Friedewald formula.14 Non-HDL cholesterol (non-HDLc) was calculated by the difference between TC and HDLc. We also calculated the ratio of non-HDLc to HDLc (non-HDLc/HDLc) and of TGs to HDLc (TG/HDLc). For the measurement of lipoprotein particles, nuclear magnetic resonance spectroscopy was used.15 The concentrations (in nmol/L) of the following particles were used according to their size (values in parenthesis) following the MESA protocol: small HDL particles (small HDLp) (7.3–8.1 nm), medium HDLp (8.2–9.3 nm), large HDLp (9.4–14 nm), small LDLp (18–20.4 nm), large LDLp (20.5–22.9 nm), intermediate dense LDLp (23–28.9 nm), small very LDL (small VLDLp) (29–34.9 nm), medium VLDLp (35–60 nm) and large VLDLp (>60 nm).

Dyslipidaemic phenotypes

To account for the strong correlation of traditional lipoproteins and lipoprotein particles and to allow for determining the main lipid phenotypes present in individuals with and without CKD, principal component analysis was used.16 This method is efficient in reducing the original number of correlated variables to a set of uncorrelated components containing most of the variance present in the original variables. Previous studies that have used principal components to account for the extensive correlation of lipoproteins and lipoprotein particles have identified two well-defined independent axes of lipid-related cardiovascular risk: the first axis characterised by the clustering of TG-rich lipoproteins and the second axis characterised by cholesterol-rich lipoproteins.17

Coronary heart disease

In MESA, identification of clinical events was done through active surveillance by trained staff. If a possible CHD event was identified, it was adjudicated as such if meeting the following criteria10: (1) myocardial infarction (cardiac biomarker elevation two times the upper limit of normal or a combination of chest pain, ECG changes and elevated cardiac biomarker 1–2 times the upper limit of normal); (2) new diagnosis of angina if typical anginal pain with a physician diagnosis and medical treatment, or if followed by revascularization; (3) resuscitated cardiac arrest if successfully recovered from cardiac arrest through cardiopulmonary resuscitation including cardioversion or (4) a CHD-related death.

Statistical methods

Demographic and clinical characteristics of the study population were compared using non-parametric statistics for continuous variables and the Χ2 statistic for categorical variables in individuals with and without CKD. Generalised linear models assuming a Gaussian distribution and an identity link function were used to obtain adjusted concentrations of lipids and lipoprotein particles with a robust variance estimation for non-normally distributed variables.

Cox-proportional hazard models were used to estimate the longitudinal associations between lipoproteins, lipoprotein particles and principal components with incident CHD. Predictors were modelled continuously (scaled to their SD or per unit increase in log-transformed variables) and categorically in their association with incident CHD. The Harrell’s C-statistic was used to determine the overall discriminative power of different models including principal components relating to the development of incident CHD. Likelihood ratio tests were used to compare different models and interaction terms in individuals with and without CKD.

All models were adjusted for age, gender, race, the presence of hypertension (present if average BP ≥140 systolic or ≥90 diastolic on the last two readings taken 5 min apart—out of three readings—or, if on antihypertensive medications), the presence of diabetes (fasting glucose ≥126 mg/dL or if on treatment with either oral hypoglycaemic agents or insulin), albuminuria (urinary albumin to creatinine ratio >30 mg/g), tobacco use, continuous log-transformed body mass index and current use of any lipid-lowering therapy including statins.

For the modelling of principal components, lipids, lipoproteins and lipoprotein particles with CHD incidence, no adjustment for other lipid components was used. An exception to this was for the analyses of TGs and HDLc and for the analyses of large and small LDLp. Given the known inverse correlation that exists between TGs and HDLc and between large and small LDLp, adjustment for each other was used when modelling their association with CHD following methods used on previous studies.18 19


Study population

CKD (eGFR 15 to <60 mL/min/1.73 m2) was present in 577 (9%) participants at baseline. For those with CKD, median eGFR (IQR) was 52 mL/min/1.73 m2 (45–57 mL/min/1.73 m2). Compared with those without CKD (table 1), CKD individuals were on average older, had a higher body mass index (BMI) and were more likely to be female, white and to have a higher prevalence of diabetes, hypertension, albuminuria (urinary albumin to creatinine ratio >30 mg/g) and use of a lipid-lowering therapy. The concentrations of TC, LDLc and HDLc were lower, and the concentration of TG was higher in the CKD group than in the non-CKD group.

Table 1

Demographic and clinical characteristics stratified by level of kidney function

Dyslipidaemic phenotypes

A strong correlation was present among the different lipoproteins and lipoprotein particles analysed (online Supplementary appendix 1A). Principal component analyses resulted in the first three components accounting for approximately 70% of the total variance in all lipoprotein measurements. Of these, two distinctive and independent dyslipidaemic phenotypes were observed (table 2 and online Supplementary appendix 1B): principal component 1 (PC1) with positive contributions from TGs and TG-rich lipoproteins (VLDLp), as well as smaller and denser LDL and HDL particles and principal component 2 (PC2) consisting of cholesterol-rich lipoproteins (TC, LDLc, large LDL). CKD was strongly associated with PC1 (P<0.001) but not with PC2 (P=0.1).

Supplemental material

Table 2

Pairwise correlation coefficients* between PC1 and PC2 with individual lipids, lipoproteins and lipoprotein particles

Lipids, lipoproteins and lipoprotein particles

Mean concentrations (adjusted to the mean values of continuous variables) of principal components and lipids and lipoprotein particles are shown in table 3. Mean adjusted values of PC1 were higher in the CKD group compared with the non-CKD group. In addition, adjusted concentrations of TG, VLDLp, sdLDL were higher in CKD, while adjusted HDLc and HDLp concentration (in particular large HDL) were lower. In contrast, PC2, TC, LDLc and large LDL were not significantly associated with CKD and tended to have a reduced (although not significant) concentration in individuals with CKD when compared with those without CKD. The non-HDLc/HDLc and TG/HDLc (but not non-HDLc) ratios were significantly associated with CKD.

Table 3

Adjusted* mean lipid, lipoproteins and lipoprotein particle concentrations stratified by CKD (CKD defined by eGFR <60 mL/min/1.73 m2)

Incident CHD and CHD prediction

Incident CHD developed in 72 individuals (12%) with CKD as compared with 303 individuals (5%) without CKD. Unadjusted incident CHD event rates for CKD individuals were 18.2 per 1000 person-years and 6.5 per 1000 person-years in those without CKD (p for difference <0.001).

Each component of PC1 was associated with CHD events in those with CKD. Specifically, elevated VLDLp concentrations (figure 1) and lower large HDLp concentrations (online Supplementary appendix 2A) were associated with a higher risk of incident CHD in those with CKD compared with those without CKD (P for multiplicative interaction <0.05). Similarly, small-LDLp was strongly associated with CHD in those with CKD (P<0.01). Accounting for the strong correlation among these lipoproteins, PC1 as a whole was a strong independent predictor of incident CHD in those with CKD (figure 1), with a 51% higher risk (per SD increase in PC1) of developing a new CHD event, compared with 13% (per SD increase in PC1) in those without CKD (P for multiplicative interaction=0.05).

Figure 1
Figure 1

HR* and 95% CI for the adjusted** association of lipids, lipoproteins, lipoprotein particles and principal components (per one SD increase in predictor except for HDLc and HDLp where estimates are per one SD decrease in predictor) and coronary heart disease incidence, stratified by CKD (CKD defined by eGFR <60 mL/min/1.73 m2). *HRs per one SD increase in predictor (values in parenthesis): principal component 1 (PC1=2.15 units), log-transformed triglycerides (log-TG per unit increase), high-density lipoprotein cholesterol (HDLc per SD decrease=15 mg/dL), log-transformed very low-density particles (log-VLDp per unit increase), small low-density lipoprotein particles (small-LDLp=370 nmol/L), high-density lipoprotein particles (HDLp per SD decrease=6.63 nmol/L), principal component 2 (PC2=1.64 units), total cholesterol (TC=35 mg/dL), low-density lipoprotein cholesterol (LDLc=31 mg/dL), low-density lipoprotein particles (LDLp=335 nmol/L), large low-density lipoprotein particles (large-LDLp=250 nmol/L), non-high-density lipoprotein cholesterol (non-HDLc=34 mg/dL), non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio (non-HDLc/HDLc =1.18 units), triglycerides to high-density lipoprotein cholesterol ratio (TG/HDLc=1.98 units). **Adjusted for age, gender, race, hypertension, diabetes, log-transformed body mass index, albuminuria (urinary to albumin creatinine ratio >30 mg/g), tobacco use and current use of lipid lowering therapy. Estimates for TG and HDLc adjusted for each other. Estimates for large-LDLp and small-LDLp adjusted for each other.

In contrast, the cholesterol-predominant phenotype (PC2) and each of its individual components including TC, LDLc and large LDLc were not associated with significantly higher risk of incident CHD in those with CKD. Individual estimates of the association between each lipoprotein particle concentration and CHD incidence is shown in (online Supplementary appendix 2A).

In risk-prediction models, the addition of PC1 to other known cardiovascular risk factors significantly improved the discrimination of the model used to predict incident CHD in those with CKD (C-statistic=0.68 in models with PC1 vs 0.66 in models without PC1; P for difference <0.01) (table 4). Compared with PC1, similar improvements in prediction of incident CHD were obtained when adding the non-HDLc/HDLc ratio and TG/HDLc ratio in participants with CKD. In contrast, no improvements in prediction for incident CHD were observed when adding PC2, TC and LDLc to the models in those with CKD.

Table 4

Predictive models for the association of traditional cardiovascular risk factors* and added lipids, lipoproteins, principal components and lipoprotein ratios with the association of coronary heart disease incidence

The proportionality assumption of the Cox models was evaluated with the use of Schoenfeld residuals with P values associated with variable specific and global tests greater than 0.05.

Sensitivity analyses

The results shown above were robust to different inclusion and exclusion criteria for the study population and to different modelling strategies. Specifically, results were consistent when excluding individuals on lipid-lowering therapy (online Supplementary appendix 3), when excluding TC, LDLc and large-LDLp from the analyses of PC1 with CHD incidence, when substituting BMI for waist circumference, and when analysing albuminuria (urinary to albumin creatinine ratio (UACR)) as a continuous variable. Although risk estimates for the association of HDLp and HDL subfractions with CHD were attenuated when using a combined cystatin C and creatinine-based formulas for GFR estimation (online Supplementary appendix 4), the order of magnitude of risk estimates were not substantially changed. Given that the Friedewald equation (18) was used for LDL estimation, in 86 individuals (<1% of the total study population) with TG levels >400 mg/dL, LDLc was unable to be estimated. Excluding these individuals from the analyses did not significantly change the main results. Results were also consistent when using a more specific definition for CHD (combined myocardial infarction and CHD-related death), when excluding participants with albuminuria (n=640), when excluding participants with eGFR <30 mL/min/1.73 m2 (n=26) and when accounting for the competing risk of death and of incident heart failure.


This study demonstrated that in individuals with stages 3 and 4 CKD, a dominant dyslipidaemic phenotype consisting of higher concentrations of TG-rich lipoproteins and other closely correlated lipoprotein abnormalities is dominant (over a cholesterol-rich phenotype) and strongly associated with CHD risk in a large and racially diverse cohort. Through the use of a complete lipoprotein assessment, we build on our previous research that found significant associations between alterations in these lipoproteins and subclinical atherosclerosis8 9 to show the importance of this phenotype on CHD in individuals with CKD.

We show that the dominant lipid phenotype in individuals with CKD consists of high VLDLp and small dense LDLp concentrations and lower concentrations of HDLp (in particular, low large HDLp concentrations). Through principal component analyses, the strong correlation that exists between these particles was accounted for and it was shown that this distinctive phenotype is strongly associated with CHD risk in CKD with a possible heterogeneity of effect by disease state (CKD). Although elevations in TGs and reductions in HDLc lipoproteins have long been described in CKD and ESRD,20 we are not aware of any previous study evaluating the association of a full spectrum of lipoprotein abnormalities with CHD in this population.

The biological mechanisms that may cause individuals with CKD to differentially express this phenotype are not entirely known. It has been shown that individuals with CKD have reduced lipoprotein lipase concentrations, reduced expression and activity of lecithin-cholesterol acyltransferase and increased activity of cholesterol ester transfer protein,21 which may account for the observed higher TG and lower HDLc concentrations in CKD. In addition, individuals with moderate CKD have been shown to have altered metabolism of apolipoprotein C-III leading to high levels of VLDL independent of the presence of diabetes.22 These typical lipoprotein abnormalities that as we show appear early in the course of CKD become more pronounced with disease severity with the great majority of patients with ESRD expressing abnormalities in TGs and HDL concentration and composition but normal or low concentrations of TC and LDLc.23

Supporting a potential causal role of CKD in the expression of this TG-rich lipoprotein phenotype, is the fact that children with CKD, who unlike adults are more commonly afflicted by primary CKD with few additional comorbidities, have a high burden of this dyslipidaemic phenotype.24 A potential unifying explanation for the observed dyslipidaemic pattern of CKD is that of insulin resistance which has been shown to be an early metabolic alteration in CKD and independent of diabetes and obesity in this population.25 More research is, however, needed to further the understanding of the likely complex relationship between kidney function, insulin resistance and lipoprotein metabolism.

We are not aware of any trials done in individuals with CKD that have specifically targeted the main dyslipidaemic phenotype present in these patients, which, as demonstrated in this study, is more predictive of risk when compared with cholesterol-rich lipoproteins. In this regard, a large meta-analysis suggests a protective effect of fibrates in reducing the risk of CVD in individuals with CKD likely related to its reduction in TG concentrations.26 Most of the studies included in this meta-analysis were, however, post hoc analyses of randomised clinical trials, which excluded individuals with severe CKD and ESRD.

We found a non-significant association of TC, LDLc and a cholesterol-rich phenotype (PC2) with CHD in participants with CKD. Furthermore, no improvement in risk prediction for CHD was observed when adding TC, LDLc or PC2 to other cardiovascular risk factors in those with CKD. These results are in agreement with studies showing a lack of association of LDLc with CVD in those with CKD27 and with a meta-analyses of clinical trials showing a relative attenuation in the protective effect of statins with decreasing kidney function.28

Therapies aimed at reducing LDLc concentrations in CKD and ESRD individuals have been inconsistent in their effect at decreasing CVD risk. Trials done in ESRD25 29 failed to show a significant reduction in CVD outcomes. The Study of Heart and Renal Protection trial30 showed a significant reduction in CVD events in those with moderate to severe CKD. The protective effect of statins appeared larger at higher baseline TC and LDL concentrations suggesting greater benefit with larger absolute reductions in LDL. Our results suggest that in addition to statins, novel non-statin therapies targeting the main dyslipidaemic phenotype present in moderate CKD may prove useful in decreasing the significant cardiovascular risk experienced by this population.

Some important limitations of this study warrant further discussion. First, the relative small sample size of individuals with CKD precludes the assessment of modification of the association of the main dyslipidaemic phenotype in CKD by other clinically important covariates. Second, there were very few individuals with UACR >300 mg/g with macroalbuminuria being a known risk factor for dyslipidaemia including elevations in TC and LDLc. Therefore, these results are not applicable to individuals with macroalbuminuria and nephrotic range proteinuria. Finally, the study population consisted of individuals with mainly moderate CKD (stage 3 CKD); therefore, extrapolation of these results to individuals with more severe forms of CKD and ESRD cannot be made.

In conclusion, a distinctive dyslipidaemic phenotype consisting of TG-rich lipoproteins is the main lipid determinant of CHD risk in CKD. Future larger studies are needed to further characterise the association of this dyslipidaemic phenotype with CHD, in particular in individuals with severe CKD and in those with ESRD. If consistent with these results, therapies aimed at targeting this phenotype may prove effective in decreasing the high burden of CHD in individuals with CKD.

Key messages

What is already known about this subject?

  • Chronic kidney disease (CKD) is a major determinant of coronary heart disease (CHD) and is associated with high triglycerides and low high-density lipid cholesterol concentrations, but not with high total or low-density lipid cholesterol concentrations. The association of different dyslipidaemic phenotypes with CHD in CKD is not known.

What does this study adds?

  • We show that a characteristic lipid phenotype consisting of triglyceride-rich lipoproteins (but not cholesterol-rich ones) is the main lipid determinant of CHD in a population with moderate CKD. CHD risk was 51% higher per each SD increase in this phenotype in individuals with CKD compared with 13% in those without CKD.

How might this impact on clinical practice?

  • This study may serve to inform future non-statin trials targeting the components of the main dyslipidaemic phenotype present in this population.


We thank the staff and participants of the MESA study for their contributions. The MESA study is supported by contracts N01-HC-95159 through N01HC-95167 and N01-HC-95169 from the National Heart, Lung, and Blood Institute.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.


  • Contributors JALM, RLM, MS and IHdB planned the study. JALM and RLM analysed the data. JALM, MS and IHdB drafted the first version of the manuscript. JALM assumes full responsibility for the overall content and scientific integrity of this manuscript.

  • Competing interests None declared.

  • Ethics approval Johns Hopkins University IRB.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.