Value of Low-Density Lipoprotein Particle Number and Size as Predictors of Coronary Artery Disease in Apparently Healthy Men and Women: The EPIC-Norfolk Prospective Population Study

Participants 

We excluded all individuals who reported a history of heart attack or stroke at the baseline clinic visit. None of the cases or control subjects was on statin treatment. Cases were individuals who developed a fatal or non-fatal CAD during follow-up. For each case, 2 control subjects matched for gender, age (within 5 years), and time of enrollment (within 3 months) were identified who had remained free of CAD during follow-up.

NMR spectroscopy 

Lipoprotein subclass particle concentrations and average size of LDL particles were measured by proton NMR spectroscopy (LipoScience, Inc., Raleigh, North Carolina) as previously described (10). Particle concentrations of lipoprotein subclasses of different size were obtained directly from the measured amplitudes of their spectroscopically distinct lipid methyl group NMR signals. The following LDL subclasses were defined: intermediate-density lipoprotein (IDL) (23 to 27 nm), large LDL (21.2 to 23 nm), and small LDL (18 to 21.2 nm). Low-density lipoprotein subclass particle concentrations are given in units of nmol/l. Summation of the LDL subclass levels provides total LDL (including intermediate-density lipoprotein) particle concentrations. Weighted-average LDL particle sizes (in nm diameter units) are computed as the sum of the diameter of each subclass multiplied by its relative mass percentage as estimated from the amplitude of its methyl NMR signal. Low-density lipoprotein subclass distributions determined by NMR and gradient gel electrophoresis are highly correlated (12). Low-density lipoprotein subclass diameters, which are consistent with electron microscopy data (13), are uniformly ∼5 nm smaller than those estimated by gradient gel electrophoresis.

Biochemical analyses 

Serum levels of total cholesterol, HDL-C, and triglycerides were measured with the RA 1000 (Bayer Diagnostics, Basingstoke, United Kingdom), and LDL-C levels were calculated with the Friedewald formula (14). Nuclear magnetic resonance analysis was performed on stored serum samples that were analyzed in random order to avoid systemic bias. Researchers and laboratory personnel were blinded to identifiable information, and could identify samples by number only.

Statistical analysis 

Baseline characteristics were compared between cases and controls taking into account the matching. A mixed effect model was used for continuous variables, and conditional logistic regression was used for categoric variables. Because triglyceride levels had a skewed distribution, values were log-transformed before being used as continuous variables in statistical analyses.

Spearman correlation coefficients and corresponding p values were calculated to assess associations between the various biomarkers and established continuous CAD risk factors. To assess the strength of association between a risk factor and the occurrence of CAD, we calculated odds ratios (ORs) and corresponding 95% confidence intervals (CIs) by conditional logistic regression analysis, taking into account matching for gender, age, and enrollment time. The ORs were calculated per quartile of each risk factor, with the first quartile as the referent group. P values represent significance for linearity across the ORs for the 4 quartiles of each risk factor. To compare the individual strengths of disease association of LDL-P, non–HDL-C, LDL-C, HDL-C, and triglycerides, we calculated ORs for future CAD per quartile of each variable in separate models adjusted for smoking (yes/no) and systolic blood pressure. Since our objective was to determine relations of lipids/lipoproteins with CAD, we did not adjust additionally for body mass index (BMI) and diabetes, 2 lipid-altering risk factors. Multivariable models were also examined to determine how relations of each variable were affected by adjustment for the other lipid/lipoprotein variables.

We further assessed the relation of LDL-P and non–HDL-C with future CAD using a model that included the FRS. We calculated this score using a previously reported algorithm (15) based on age, gender, levels of total and HDL cholesterol, systolic and diastolic blood pressure, diabetes, and smoking and categorized subjects into 3 groups: low (<10%), intermediate (10% to 20%), or high (>20%) risk. The ORs for future CAD were calculated per quartile of the risk factor, adjusting for the FRS category.

Statistical analyses were performed using SPSS software (version 12.0.1, SPSS Inc., Chicago, Illinois). A p value <0.05 was considered to indicate statistical significance.

Baseline characteristics 

We identified 1,003 participants who were apparently healthy at baseline and developed CAD during follow-up. A total of 882 cases were matched to 2 controls each, whereas the remaining 121 cases could be matched to 1 control only, giving a total number of 1,885 controls. Baseline characteristics of cases and controls are listed in Table 1. As expected, cases were more likely to be smokers and diabetic subjects, and to have a higher blood pressure and BMI than control subjects. Levels of total cholesterol, LDL-C, non–HDL-C, and triglycerides were significantly higher in cases than in control subjects, whereas HDL-C levels were significantly lower (p < 0.0001 for each). Baseline LDL-P was higher in cases compared with control subjects (p < 0.0001). Levels of the large LDL subclass were not different, but cases had more IDL and small LDL particles. Thus, the increased LDL-C in cases is attributable mainly to increased cholesterol in small LDL. The average LDL particle size was smaller in cases than control subjects.

Table 1. Baseline Characteristics of Coronary Artery Disease Cases and Matched Control Subjects

		Control Subjects (n = 1,885)	Cases (n = 1,003)	p Value
	Male gender, % (n)	63.2(1,192)	63.9(641)	Matched
	Age, yrs	65±8	65±8	Matched
	Diabetes, % (n)	1.6(30)	6.1(61)	<0.0001
	Smoking, % (n)	8.3(155)	15.7(157)	<0.0001
	BMI, kg/m2	26.2±3.4	27.3±3.9	<0.0001
	Systolic blood pressure, mm Hg	139±18	144±19	<0.0001
	Diastolic blood pressure, mm Hg	84±11	86±12	<0.0001
	Chemical lipid measures
	Total cholesterol, mmol/l	6.2(5.5–6.9)	6.4(5.6–7.2)	<0.0001
	HDL cholesterol, mmol/l	1.3(1.1–1.6)	1.2(1.0–1.8)	<0.0001
	LDL cholesterol, mmol/l	4.0(3.4–4.7)	4.2(3.6–4.9)	<0.0001
	Triglycerides, mmol/l	1.6(1.1–2.2)	1.8(1.3–2.6)	<0.0001
	Non-HDL cholesterol, mmol/l	4.8(4.1–5.6)	5.2(4.4–5.9)	<0.0001
	NMR LDL particle measures
	LDL particle number, nmol/l	1,525(1,278–1,812)	1,640(1,383–1,955)	<0.0001
	IDL, nmol/l	36(14–66)	43(19–78)	0.003
	Large LDL, nmol/l	572(448–704)	568(427–708)	0.6
	Small LDL, nmol/l	885(637–1,190)	999(747–1,330)	<0.0001
	LDL size, nm	21.1(20.7–21.5)	21.0(20.1–21.4)	0.002

Data are presented as mean ± SD, percentage (n), or median (interquartile range). Means, percentages, and medians may be based on fewer observations than the indicated number of subjects. Triglyceride levels were log-transformed before analysis. BMI = body mass index; HDL = high-density lipoprotein; IDL = intermediate-density lipoprotein; LDL = low-density lipoprotein; NMR = nuclear magnetic resonance.

Associations of LDL measures with other CAD risk factors 

As shown in Table 2, LDL size was inversely correlated with LDL-P (r = −0.58), but not with LDL-C (r = 0.01). We identified a strong inverse relation of LDL size with triglyceride levels (r = −0.53) and BMI (r = −0.19) and a positive correlation with HDL-C (r = 0.54). Both LDL-C and LDL-P were strongly interrelated (r = 0.63), but both parameters had markedly different associations with HDL-C and triglycerides. Whereas LDL-C was only weakly associated with HDL-C (r = −0.03) and triglyceride levels (r = 0.18), LDL-P was more strongly correlated with HDL-C (r = −0.29) and triglycerides (r = 0.55). Non–HDL-C was strongly correlated with LDL-P (r = 0.76) and triglycerides (r = 0.47), and inversely correlated with HDL-C (r = −0.16) and LDL size (r = −0.18).

Table 2. Spearman Correlation Coefficients Between Measured Variables⁎

		Non-HDL Cholesterol	LDL Cholesterol	LDL Particle Number	LDL Size
	LDL cholesterol	0.94
	LDL particle number	0.76	0.63
	LDL size	−0.18	0.01†	−0.58
	Total cholesterol	0.94	0.93	0.65
	HDL cholesterol	−0.16	−0.03‡	−0.29	0.54
	Triglycerides	0.47	0.18	0.55	−0.53
	BMI	0.11	0.02§	0.14	−0.19

BMI = body mass index; HDL = high-density lipoprotein; LDL = low-density lipoprotein.

⁎ p < 0.0001 for comparison unless otherwise indicated † p = 0.6 ‡ p = 0.09 § p = 0.2.

LDL-P, traditional lipid risk factors, and risk for future CAD 

Shown in the upper panel of Table 3 are the univariate ORs for future CAD associated with increasing quartiles of non–HDL-C, LDL-C, LDL-P, HDL-C, and triglycerides. The 5 lipid/lipoprotein measures exhibited comparable strengths of association with CAD, with ORs differing approximately 2-fold comparing the highest and lowest quartiles. The magnitude of the predictive value of LDL-P and non–HDL-C were greater than that of LDL-C. Comparing the highest to lowest quartile, the OR for LDL-P was 2.00 (95% CI 1.58 to 2.59) and 2.14 (95% CI 1.69 to 2.69) for non–HDL-C versus 1.73 (95% CI 1.37 to 2.18) for LDL-C.

Table 3. Odds Ratios for Future Coronary Artery Disease by Quartile of Lipid/Lipoprotein Variable in Univariable and Multivariable Models⁎

		Quartiles				p Value†
		1	2	3	4
	Univariable models
	LDL cholesterol	1.00	1.37(1.09–1.73)	1.38(1.09–1.74)	1.73(1.37–2.18)	<0.0001
	LDL particle number	1.00	1.23(0.97–1.56)	1.48(1.17–1.87)	2.00(1.58–2.59)	<0.0001
	HDL cholesterol	1.00	0.76(0.60–0.95)	0.60(0.48–0.75)	0.50(0.39–0.64)	<0.0001
	Triglycerides	1.00	1.27(1.00–1.62)	1.50(1.20–1.88)	2.01(1.61–2.51)	<0.0001
	Non-HDL cholesterol	1.00	1.31(1.04–1.66)	1.57(1.25–1.97)	2.14(1.69–2.69)	<0.0001
	Multivariable models
	LDL cholesterol	1.00	1.26(0.99–1.60)	1.27(1.00–1.61)	1.55(1.22–1.96)	0.001
	HDL cholesterol	1.00	0.83(0.66–1.04)	0.71(0.56–0.89)	0.66(0.50–0.87)	0.001
	Triglycerides	1.00	1.12(0.88–1.43)	1.22(0.96–1.54)	1.52(1.19–1.95)	0.001
	LDL particle number	1.00	1.13(0.89–1.44)	1.21(0.94–1.54)	1.37(1.04–1.83)	0.02
	HDL cholesterol	1.00	0.86(0.68–1.09)	0.74(0.59–0.94)	0.70(0.53–0.92)	0.005
	Triglycerides	1.00	1.11(0.87–1.42)	1.19(0.93–1.51)	1.36(1.04–1.79)	0.03
	Non-HDL cholesterol	1.00	1.13(0.88–1.44)	1.26(0.99–1.62)	1.63(1.26–2.11)	<0.0001
	HDL cholesterol	1.00	0.83(0.66–1.05)	0.71(0.56–0.90)	0.66(0.50–0.87)	0.001
	Triglycerides	1.00	1.08(0.85–1.39)	1.13(0.88–1.43)	1.30(0.99–1.70)	0.06

Abbreviations as in Table 2.

⁎ Odds ratios (95% confidence intervals) were calculated by conditional logistic regression, taking into account matching for gender, age, and enrollment time and adjusted additionally for smoking and systolic blood pressure. Univariable models examined each lipid/lipoprotein variable in a separate model. Multivariable models examined each variable in a model adjusted for the other 2 lipid/lipoprotein variables as continuous variables. † p for linear trend.

The lower panels of Table 3 show the results of multivariate analyses in which each lipid/lipoprotein variable was adjusted for the other 2 parameters to assess the independence of their mutual relations with CAD. The associations of both LDL-C and LDL-P with CAD were attenuated after adjustment for HDL-C and triglycerides, whereas attenuation was more pronounced for LDL-P than for LDL-C (4th quartile ORs reduced from 2.00 to 1.37 for LDL-P vs. 1.73 to 1.55 for LDL-C). Similarly, relations of HDL-C and triglycerides with CAD were attenuated more by adjustment for LDL-P than LDL-C.

Concordance/discordance between LDL-C and LDL-P 

In a conditional logistic regression model that included both parameters and corrected for smoking and systolic blood pressure, LDL-C was no longer statistically significantly associated with future CAD (Table 4). The LDL-P, however, remained a significant risk factor, such that subjects in the highest quartile had an OR of 1.78 (95% CI, 1.34 to 2.37; p < 0.0001 for linearity) (Table 4).

Table 4. Odds Ratios for Future Coronary Artery Disease by Quartile of LDL-C and LDL-P, Both Entered Into One Model⁎

		LDL Size Quartile
		1	2	3	4	p Value†
	LDL-C	1.00	1.23(0.97–1.57)	1.13(0.88–1.45)	1.22(0.92–1.61)	0.3
	LDL-P	1.00	1.18(0.93–1.51)	1.39(1.08–1.79)	1.78(1.34–2.37)	<0.0001

LDL-C = low-density lipoprotein cholesterol; LDL-P = low-density lipoprotein particle number.

⁎ Odds ratios (95% confidence intervals) were calculated by conditional logistic regression adjusted for smoking and systolic blood pressure † p for linear trend.

LDL size and risk for future CAD 

Table 5 shows the ORs for future CAD associated with increasing quartiles of LDL size adjusted for smoking and systolic blood pressure, with and without additional adjustment for the potentially confounding inverse correlation of LDL size with LDL-P. Without adjustment for LDL-P, there was a significant relation of smaller LDL size with CAD, with an OR for individuals in the highest quartile compared with those in the lowest quartile of 0.60 (95% CI 0.47 to 0.76). However, upon adjustment for LDL-P, the relation of LDL size with CAD was greatly attenuated and was no longer significant.

Table 5. Odds Ratios for Future Coronary Artery Disease by Quartile of LDL Size, With and Without Adjustment for LDL Particle Number⁎

	Range(nm)	LDL Size Quartile				p Value†
		1	2	3	4
		<20.6	20.7–21.0	21.1–21.4	>21.4
	Unadjusted for LDL particle number	1.00	0.77(0.62–0.97)	0.76(0.60–0.95)	0.60(0.47–0.76)	<0.0001
	Adjusted for LDL particle number	1.00	0.92(0.72–1.16)	0.99(0.77–1.28)	0.86(0.65–1.15)	0.5

⁎ Odds ratios (95% confidence intervals) were calculated by conditional logistic regression adjusted for smoking and systolic blood pressure, with and without additional adjustment for low-density lipoprotein (LDL) particle number † p for linear trend.

LDL-P and the FRS 

The results in Table 6 indicate that LDL-P and non–HDL-C retain a similar association with future CAD after accounting for the FRS (OR of 1.34 and 1.38, respectively, in the highest vs. lowest quartile).

Table 6. Odds Ratios for Future Coronary Artery Disease After Taking Into Account the Framingham Risk Score⁎

		Quartiles				p Value†
		1	2	3	4
	LDL cholesterol	1.00	1.17(0.93–1.48)	1.09(0.86–1.39)	1.24(0.97–1.58)	0.15
	LDL particle number	1.00	1.10(0.86–1.39)	1.22(0.96–1.54)	1.34(1.03–1.73)	0.02
	Non-HDL cholesterol	1.00	0.98(0.73–1.32)	1.03(0.76–1.38)	1.38(1.01–1.90)	0.04

Abbreviations as in Table 2.

⁎ Odds ratios were calculated by conditional logistic regression, with adjustment for risk categories based on the Framingham risk score † p for linear trend.

LDL-P and risk for future CAD 

The cholesterol content of LDL particles, which can create discordance between levels of LDL-C and LDL-P, is mainly influenced by cholesterol ester transfer protein activity, which is enhanced under circumstances of hypertriglyceridemia (8). This has 2 important consequences. First, transfer of cholesteryl ester from HDL particles to apoB-containing lipoproteins causes low HDL-C levels. Second, cholesterol depletion and triglyceride enrichment of LDL particles facilitates the formation of small dense LDL particles (3, 8). As a result, LDL-C levels generally underestimate the number of LDL particles in individuals with elevated triglycerides, as clearly illustrated in the Framingham Study (2, 3). Discordance between LDL-C and LDL-P is also a feature of diabetic patients (2, 16) and the number of metabolic syndrome components (2, 17). Our data in the EPIC-Norfolk study showing that triglyceride and HDL-C levels are much more strongly correlated with LDL-P than LDL-C are in agreement with these findings. Accordingly, it was not unexpected that the relation of LDL-P with CAD risk was to be weakened more than that of LDL-C by multivariate adjustment for triglycerides and HDL-C. Conversely, adjusting for LDL-P weakened relations of CAD with triglycerides and HDL-C, suggesting that some of the risk associated with these non-LDL risk factors may actually stem from elevations of LDL-P not reflected by levels of LDL-C.

We found a moderate degree of discordance between LDL-C and LDL-P in our study population. While the source of this excess risk may not be due entirely to the elevated LDL-P, since those with discordantly high LDL-P often have increased triglycerides and decreased HDL-C, it is biologically plausible that LDL-P makes a contribution. Current understanding of the pathophysiology of atherosclerotic vascular disease is that LDL particles are active participants from the time they enter the artery wall, are retained in the intima by binding to extracellular matrix, become chemically modified by oxidation, and are subsequently ingested by macrophages to create foam cells and increased plaque burden (2, 4).

Despite the finding that LDL-P predicted CAD independently of the FRS, our results, showing a loss of discriminative power of LDL-P over LDL-C when HDL-C and triglyceride levels are accounted for, do not argue for the routine implementation of LDL-P assessment for CAD risk assessment. Low-density lipoprotein cholesterol particle number may, however, play a useful role in patient management by helping judge the adequacy of LDL lowering therapy, particularly among those with elevated triglycerides and reduced HDL-C. Such a role has been proposed for apoB (4, 5), and both non–HDL-C and apoB have been put forward as secondary treatment targets after LDL-C goals have been achieved (1, 4). In fact, data supporting LDL-P as an alternative treatment target has gained support from clinical intervention studies showing that on-treatment levels of apoB or NMR-measured LDL-P were more reliable indicators of residual CAD risk than on-treatment LDL-C (2, 4, 5, 18, 19).

LDL size and risk for future CAD 

Small LDL size is another factor associated with high triglycerides, low HDL-C, obesity, insulin resistance, diabetes, and metabolic syndrome (6, 7, 8, 9). Our finding of a relation between smaller LDL size and greater CAD risk is consistent with previous studies reporting that small LDL particles have higher atherogenic potential than large LDL particles (8, 9). However, upon adjustment for LDL-P, the relationship of LDL size with CAD was abolished. This result is consistent with findings in the Multi-Ethnic Study of Atherosclerosis indicating that NMR-measured numbers of small and large LDL particles are related similarly to carotid atherosclerosis (20).

Non–HDL-C and risk for future CAD 

The present findings confirm that non–HDL-C is a better predictor of CAD than LDL-C (21, 22). Previous studies have also found that the number of most atherogenic lipoprotein particles, as measured by apoB, were more strongly related to CAD risk than was non–HDL-C (21, 22). In the present study, we show that the association of LDL-P with CAD is almost equal to that of non–HDL-C. The fact that apoB captures all atherogenic apoB particles (including very low-density lipoprotein and LDL), whereas LDL-P only measures LDL particles, may have contributed to this distinction. There has been an intense debate concerning clinical relevance of measuring particle numbers (apoB/LDL-P) and/or cholesterol content of the particles (non–HDL-C). In the present study, we observed that both LDL-P as well as non–HDL-C conferred predictive value on top of the FRS. Since the association between LDL-P and CAD was equal to that of non–HDL-C, the present findings do not advocate routine use of LDL-P in CAD risk assessment. The potential value of LDL-P measurement for monitoring patients on lipid-lowering medication needs to be addressed in separate intervention trials.

Study limitations 

Our study has several limitations. The study population was relatively elderly, which may limit the generalizability of our results. Case-control differences in CAD among older populations are more weakly related to lipoprotein levels than in younger populations because many of the control subjects have extensive subclinical disease, yet have not experienced a coronary event (23). Low-density lipoprotein cholesterol levels in the study population were also considerably higher than in the general U.S. population, with a mean LDL-C value in the EPIC-Norfolk study corresponding to the 80th percentile of Framingham subjects of similar age and gender (24). Coronary artery disease events in our study were ascertained through death certification and hospital admission data, which may lead both to under ascertainment and to misclassification of cases. Previous validation studies in this cohort, however, indicate high specificity of such case ascertainment (25).

In conclusion, in this large cohort of apparently healthy men and women, LDL-P and non–HDL-C were more closely associated than LDL cholesterol with the occurrence of future CAD. After adjustment for HDL-C and triglycerides, LDL-P was no longer more predictive than LDL-C. These findings do not support routine use of LDL-P in CAD risk assessment strategies in primary prevention strategies. However, recognition that patients with low HDL-C and/or high triglycerides often have elevated numbers of LDL particles without having elevated LDL-C may enable their LDL-related CAD risk to be managed more effectively.