Sninsky, John J. PhD*; Rowland, Charles M. MS*; Baca, Arthur M. MD†; Caulfield, Michael P. PhD‡; Superko, Harold Robert MD*
In 1913, Anitschkow and Chalatow conducted studies on rabbits that indicated high-fat diets resulted in hypercholesterolemia and subsequent arteriosclerosis.1 It is now well accepted that elevated low-density lipoprotein cholesterol (LDL-C) is a risk factor for coronary heart disease (CHD).2 However, LDL-C levels do not completely predict cardiovascular events and a recent report of 136,905 patients hospitalized with coronary artery disease for the first time revealed that 75% had an LDL-C below 130 mg/dL, the reputed goal in an asymptomatic population.3 It is worthy of note that relative risk reductions of 27.2% attributed to LDL-C lowering monotherapy actually reflect only a 3.4% absolute risk reduction.4 Thus, efforts to identify additional clinically relevant risk factors are critically important.
In the 1950s, Gofman and colleagues,5 at the Donner Laboratory (University of California), investigated the relationship of LDLs to atherosclerosis in the Framingham study and Lawrence Radiation Laboratory at Livermore study. Krauss and colleagues subsequently found that LDL is not a homogeneous category of lipoproteins but consists of a set of discrete subspecies with distinct molecular properties, including size and density.6,7 In normal subjects, 7 major LDL subfractions can be identified by ultracentrifugation, namely, I, IIa, IIb, IIIa, IIIb, IVa, and IVb. LDL-I is the largest and least dense, and LDL-IVb is the smallest and the most dense. Investigators have extended the work of Gofman and colleagues to assess the role of lipoprotein subfractions in atherosclerosis.8 These investigations bridged the gap between basic science and clinical research, and most recently have involved genetics and arteriographic trials.8–11 A number of studies have found that having a predominant fraction of small, dense LDL particles is associated with increased risk of CHD independent of LDL-C and most standard risk factors.12–16 Austin et al.12 defined an LDL phenotype as pattern A or B according to the LDL particle size determined by gradient gel electrophoresis. They noted that the proportion of pattern B was higher among CHD cases than controls. The Boston Area Health Study, the Physicians Health Survey, The Stanford Five City Project, and the Quebec Cardiovascular Study all confirmed that the abundance of small, dense LDL particles is associated with increased risk for cardiovascular events.12–17 However, it is clear that not all studies support the correlation between CHD risk and LDL particle size. A detailed review of the many studies is outside the scope of this article but comprehensive reviews of the various studies and their outcomes have been recently published.18–27 Current American Heart Association, National Lipid Association, and European Atherosclerosis Society guidelines do not recommend routine use of lipid subfraction tests.
Interest in LDL phenotype and subfraction distribution as emerging risk factors for CHD has led a number of clinical reference laboratories to offer tests that use different laboratory methods.28 In addition, the European Atherosclerosis Society29 and American Heart Association30 recently highlighted triglyceride levels (TG), high-density lipoprotein (HDL-C), and TG/HDL-C ratio as important measures of cardiovascular risk beyond LDL-C and suggested treatment strategies for patients with elevated TG or TG/HDL-C ratio or low HDL-C levels.
Particle size or density distributions, from which the LDL phenotype is determined, is measured by a variety of laboratory methods including analytical ultracentrifugation, density ultracentrifugation by vertical auto profile (VAP), segmented gradient gel electrophoresis (sGGE), tube gel electrophoresis (TGE), nuclear magnetic resonance (NMR), and ion mobility (IM). The newest of these methods is IM which works on the principle that particles of a given size and charge behave in a predictable manner when carried in a laminar flow of air and subjected to an electric field.31 Some have suggested that the TG/HDL-C ratio could be a surrogate measure of LDL particle size and phenotype.32–34 Currently, there is no standardization among the various methods and thus results from different laboratory methods may not be consistent. For example, Ensign et al.35 found complete agreement of LDL phenotype in only 8% of 40 subjects for whom LDL phenotype was measured by each of the following 4 methods: VAP, sGGE, NMR, and TGE. Such disparities may contribute to confusion regarding clinical interpretation of LDL subfraction results and limit the clinical usefulness of LDL phenotype determination.
The purpose of the current investigation is to evaluate the agreement of LDL phenotypes among IM, VAP, sGGE, and NMR in 228 subjects. Our study extends the report by Ensign et al.35 by including IM among the methods being compared, by evaluating the results on a considerably larger sample size, and determining the correlation of LDL phenotype with TG, HDL-C, and TG/HDL-C biomarkers for each of these technologies.
MATERIALS AND METHODS
Apparently healthy adult volunteers were recruited for the study. A wide spectrum of sex, age, body fatness, and lifestyle were represented. Exclusion criteria: history of fasting triglycerides greater than 1000 mg/dL that are not controlled and history of abnormal bleeding. The study design was a split sample comparison. Institutional review board exemption (Western Institutional Review Board) was obtained to use the anonymized blood samples. Blood samples were collected during July and August of 2011. Blood was collected using standard venipuncture techniques, and serum separated from red blood cells within 2 hours of collection. All samples were transported by the recommended conditions to Atherotech (Birmingham, AL) for VAP, Berkeley HeartLab (Alameda, CA) for sGGE, Liposcience (Raleigh, NC) for NMR, and Quest Diagnostics Nichols Institute (San Juan Capistrano, CA) for IM. Descriptions of the technologies were as reported in Ensign et al.35 with IM as described by Caulfield et al.31 with minor modifications. Lipoproteins were isolated using a dextran sulfate precipitation procedure that no longer required the use of the ultracentrifuge. This new procedure resulted in greater than 96% recovery of lipoproteins (as judged by TC, Apo A1, and ApoB recoveries) with greater than 96% removal of predominant plasma proteins; albumin (96%), IgG (97%), fibrinogen (97%), and transferrin (100%). Lipoproteins were recovered from the precipitate in an ammonium acetate solution and were diluted in ammonium acetate before analysis. The intra-assay and inter-assay variation (CV) was less than 10% and less than 13%, respectively, for all lipoprotein fractions. The LDL pattern for each method was assigned by LDL particle size for sGGE, NMR, and IM and based on density for VAP. VAP reports the LDL phenotype as pattern A (large), A/B (intermediate), or B (small) based on their vertical auto profile-II test which is derived from density gradient ultracentrifugation.36–38 sGGE also reports the LDL phenotype as pattern A, A/B, or B along with the peak LDL particle size (in angstroms).39–41 sGGE uses a cutoff for pattern B of less than 255 A.37 NMR reports peak LDL size (in nanometers) and LDL phenotype as pattern A or B.42 NMR uses a cutoff for pattern B of less than 20.5 nm. The NMR size cutoff is correlated to the cutoff used in sGGE.43 IM reports LDL size (in angstroms) and LDL phenotype as pattern A or B. IM uses a cut point of less than 216.0 A. This value represents the lower end of the 95th percentile of the LDL size distribution for the female population in the reference interval determination and was not determined by correlation of IM with sGGE.
TG and HDL-C were tested in 3 laboratories (Berkeley HeartLab, Atherotech, and Liposcience) and had a pairwise correlation (r) of 0.98 or greater. An average of the individual results for TG and HDL-C from 3 laboratories was used for analyzing the correlation between LDL phenotypes and TG, HDL-C, and TG/HDL-C.
Contingency tables were formed for each pair of methods and agreement was assessed using 2 metrics, namely, percent agreement and the κ statistic.44 Percent agreement was defined as the percent of subjects who received identical classifications by both methods. κ measures agreement adjusted for agreement that would be expected by chance (if classifications by the 2 methods were statistically independent). κ ranges between −1 and 1, where 1 represents perfect agreement and 0 represents chance agreement. When measures with more than 2 categories were being compared and the categories could be thought of as being ranked in an ordinal manner, κ was weighted to penalize some mistakes more severely than others. Thus, for comparison of A, A/B, and B pattern calls, we used weights that penalize disagreements that are further from the diagonal of the 3 × 3 table (classified as A by one method and B by the other method) more severely than disagreements that lie close to the diagonal (classified as A/B by one method and either A or B by the other method).45
Because the sGGE and VAP methods may result in pattern A, A/B, or B, whereas NMR and IM result only in pattern A or B, agreement was assessed in a variety of ways for the comparison of pairs of methods having different call possibilities. For such pairs of methods, percent agreement was assessed in each of 4 ways as follows: (1) use all data (agreement requires identical calls and all calls contribute to denominator, (2) exclude subjects with pattern A/B, (3) collapse patterns into A and not A (A/B or B), and (4) collapse patterns into B and not B (A/B or A). κ was assessed only for the latter 3 ways because it is not defined for non–square contingency tables.
Logistic regression models were used to evaluate the relationship between concordance of LDL phenotype among the 4 methods and TG/HDL-C ratio modeled as a continuous variable. Models including a second-order polynomial of the TG/HDL-C were also assessed to allow for the possibility of a nonlinear relationship.
Pearson correlation coefficient was used to evaluate the strength of relationships between continuous measures. Differences in correlation coefficients were assessed using the “D1” method described by Wilcox and Tian.46
Blood samples were collected from 228 volunteers, 45% of whom were male. The subjects ranged in age from 21 to 68 years with a median age of 41 years. Measures of LDL-C, HDL-C, and TG are shown in Table 1.
Agreement of LDL Phenotype Among All 4 Methods
Because of limited sample volumes in some cases, LDL phenotype was evaluated by NMR in only 190 of the 228 subjects and by IM in 202 subjects. Among the 165 subjects evaluated by all 4 methods, there was complete agreement for 106 (64.2%) subjects and agreement between 3 of the 4 methods for 86.7% of subjects (Table 2).
When subjects with pattern A/B by either VAP or sGGE were excluded, 123 subjects with either pattern A or B in any method were available for analysis. There was agreement between at least 3 methods in 99.2% of subjects and between all 4 methods in 86.2% of subjects. Of the 16 subjects with agreement on 3 of 4 methods, the discordant method was NMR for 10 subjects, VAP for 5, and IM for 1.
When patterns were collapsed into B versus not B, there was agreement between at least 3 methods in 94.5% of subjects and between all 4 methods in 73.9% of subjects. Of the 34 subjects with agreement on only 3 of 4 methods, the discordant method was NMR for 14 subjects, VAP for 12, sGGE for 5, and IM for 3.
When patterns were collapsed into A versus not A, there was agreement between at least 3 methods in 92.7% of subjects and between all 4 methods in 67.9% of subjects. Of the 41 subjects with agreement on only 3 of 4 methods, the discordant method was NMR for 13, VAP for 11, sGGE for 10, and IM for 7.
Logistic regression models which assume a linear relation between the log odds of LDL phenotype concordance (B vs not B) and TG/HDL-C ratio found that the odds of concordance decreased as TG/HDL-C ratio increased (P = 0.02) when the definition of concordance required all 4 methods to agree. When the concordance definition required at least 3 methods to agree, there was a nonsignificant downward trend (P = 0.16). However, addition of a polynomial term of (TG/HDL-C)2 significantly improved the fit of the model requiring concordance of all 4 methods (P < 0.0001) as well as the fit of the model requiring concordance among at least 3 methods (P = 0.02) and indicated a nonlinear relationship wherein concordance is lowest among subjects with intermediate values of TG/HDL-C and higher among subjects having either low or high levels of TG/HDL-C (Fig. 1). Results were similar when concordance for all categories was considered (A,A/B,B) (data not shown).
Agreement of LDL Phenotype Among Each Pair of Methods
The cross-classifications of phenotypes for each pair of methods are shown in Table 3; percent agreement and κ for the pairwise comparisons are shown in Table 4.
IM Versus sGGE
IM assigned 79% of subjects to pattern A and 76% of these subjects were also assigned pattern A by sGGE. Among subjects who were assigned B by IM, sGGE assigned 71% to pattern B, 24% to pattern A/B, and the remaining 5% to pattern A. Overall, 24% of calls were A/B by sGGE and either A or B by IM, and 1% of calls were A by one method and B by the other method (Table 3). Agreement ranged from 98% when sGGE pattern A/B subjects were excluded to 75% when all data were considered. κ ranged from 0.94 when sGGE pattern A/B subjects were excluded to 0.53 when sGGE patterns were evaluated as A versus not A (Table 4).
The 2 subjects who were pattern A by sGGE but B by IM as well as the 1 subject who was pattern B by sGGE but pattern A by IM had peak particle sizes that placed them close to the cluster of pattern A/B subjects by sGGE (Fig. 2B). Similarly, all of the pattern A/B subjects classified by sGGE had peak particle size close to the IM cut point of pattern A versus B. Two subjects classified as pattern A/B by sGGE seem to cluster among sGGE pattern B subjects; however, both had a second LDL peak size that did fall in the cluster of A/B pattern (Fig. 2B).
IM Versus VAP
Among the 160 subjects assigned to pattern A by IM, 78% were also assigned pattern A by VAP and among 42 subjects assigned to pattern B by IM, 81% were also assigned pattern B by VAP. Overall, 10% of calls were pattern A/B by VAP and either A or B by IM, and 11% of calls were pattern A by one method and pattern B by the other (Table 3). Agreement ranged from 79% when all data were considered to 87% when VAP pattern A/B subjects were excluded. κ ranged from 0.55 when VAP patterns were evaluated as A versus not A to 0.67 when VAP pattern A/B subjects were excluded (Table 4).
IM Versus NMR
Among the 130 subjects assigned to pattern A by IM, 88% were also assigned pattern A by NMR and among 35 subjects assigned to pattern B by IM, 74% were also assigned pattern B by NMR (Table 3). The methods agreed on 85% of calls and κ was 0.58 (Table 4). LDL size among the subjects with discordant results was primarily clustered near the pattern A and pattern B thresholds (Fig. 2C).
sGGE Versus VAP
VAP assigned a higher proportion of subjects to pattern B than did sGGE (27% and 15%, respectively). The methods agreed on 73% of subjects, whereas 24% of subjects were pattern A/B by one method and either pattern A or pattern B by the other method, and 8 (4%) of results were pattern A by one method and pattern B by the other method. Of the 8 most discrepant results (A vs B discrepancies), 7 were pattern B by VAP and pattern A by sGGE (Table 3). κ was 0.72 (Table 4).
sGGE Versus NMR
NMR assigned 74% of subjects to pattern A and 83% of these subjects were also assigned pattern A by sGGE. Agreement was not as high for subjects who were pattern B by NMR with only 50% of pattern B subjects by sGGE. Overall, 18% of subjects were pattern A/B by sGGE and either pattern A or B by NMR, and 8% of subjects were pattern A by one method and B by the other (Table 3). Agreement ranged from 74% when all data were considered and 90% when pattern A/B results were excluded. κ ranged from 0.53 when patterns were evaluated as B versus not B to 0.71 when pattern A/B results were excluded (Table 4).
An LDL peak size greater than or equal to 20.6 nm discriminated pattern A from pattern B for the NMR method. Several, but not all, of the subjects with discordant patterns (A vs B) by the 2 methods had LDL peak sizes close to the threshold (Fig. 2A).
VAP Versus NMR
Among the 140 pattern A subjects by NMR, 81% were also pattern A by VAP. Among 51 pattern B subjects by NMR, 65% were pattern B by VAP. Overall, 8% of the subjects were pattern A/B by VAP and either A or B by NMR, and 15% were pattern A by one method and pattern B by the other method (Table 3). Agreement ranged from 77% when all data were considered and 84% when pattern A/B subjects were excluded. κ ranged from 0.53 when patterns were evaluated either as A versus not A or B versus not B to 0.58 when pattern A/B subjects were excluded (Table 4).
Correlations of TG, HDL-C, and TG/HDL-C Ratio With Peak LDL-C Size
Absolute values of Pearson correlation coefficients correlating TG and HDL-C values, and TG/HDL-C ratio with LDL peak size determined from sGGE, NMR, and IM among the 165 subjects having results for all methods ranged from 0.58 to 0.78. High TG and TG/HDL-C ratio were correlated with small LDL size. Lower HDL-C was also correlated with small peak LDL size. The strength of correlation between TG/HDL-C ratio and LDL size were −0.78 for IM, −0.69 for sGGE, and −0.64 for NMR. The differences in strength of correlation were modest between IM and sGGE (P = 0.048) and significant between IM and NMR (P = 0.005). Similar results were found for correlations of TG with LDL size. There was no difference (P > 0.2) in strength of correlation between HDL-C and the peak LDL size as measured by the 3 methods (Table 5).
In this study, we evaluated the agreement between IM, VAP, sGGE, and NMR among 228 subjects and found that the 4 methods of LDL phenotype determination had complete agreement among 64% of subjects and agreement of at least 3 of 4 methods for 87% of subjects. Excluding subjects with pattern A/B by VAP or sGGE, there was agreement between at least 3 methods in 99% of subjects and between all 4 methods in 86% of subjects.
The findings of this study showed considerably better agreement among methods than the 8% complete agreement among TGE, sGGE, VAP, and NMR reported by Ensign et al.35 And in contrast to the prior report, the pairwise correlations of particle size between the various methods showed good correlation in this study. In both this study and the prior study, if samples with pattern A/B were excluded, there was 90% agreement between sGGE and NMR. The high concordance between the methods investigated in this study may be in part due to the exclusion of the TGE method, and because IM had better agreement with VAP, NMR, and sGGE than does TGE. This study also has a larger sample size than the previous study, and thus provides a better estimate of the agreement between methods. It is also possible that the clinical reference laboratories have refined their procedures in various ways since the study of Ensign et al.35 was performed.
Comparison between pairs of methods was limited by the fact that VAP and sGGE report an intermediate A/B pattern, whereas IM and NMR do not. Our finding that VAP reported a higher proportion of pattern B calls than sGGE confirmed the trend reported by Ensign et al.35 When pattern A/B subjects were excluded, pairwise agreement was highest between sGGE and IM, with only 3 discordant subjects. In addition, if sGGE patterns were evaluated as B versus not B, sGGE and IM had 94% agreement. Although LDL peak diameters determined by the 4 methods do not correspond in size, the strong agreement between sGGE and IM for pattern B results suggest that the sGGE cut point between pattern A/B and B and the single IM cut point between pattern A and pattern B are nearly equivalent.
We confirmed previous reports that TG, HDL-C, and TG/HDL-C ratio are correlated with peak LDL size.33,34 This correlation was stronger when peak LDL size was determined by IM than by sGGE or NMR. We also found that concordance among the 4 methods tends to be lower among subjects with intermediate levels of TG/HDL-C ratio. This study has several limitations. First, the study was performed on volunteer subjects who may not be representative of the population of patients for whom these tests would typically be performed. However, the distribution of LDL phenotypes as well as TG and HDL-C values, and TG/HDL-C ratio were similar to preventative cardiology outpatient populations. Second, the lack of standardization in reporting makes it challenging to directly compare the methods. Specifically, the algorithm used to determine LDL phenotype for some methods is not explicitly reported. Therefore, the difference between methods could be due to measurement errors as well as differences between the criteria used to report LDL phenotype. Finally, the subjects in this study do not have clinical outcomes so we were unable to assess the correlation between LDL phenotypes and clinical outcomes.
The findings of this study have shown that agreement between methods for determining LDL phenotype is better than what has been reported previously.35 The methods have complete agreement for approximately two thirds of subjects, and approximately 75% of subjects are in agreement for any pair of methods. The results of the current study provide more reliable estimates of pairwise agreement given that the sample size was approximately 4 times larger than that of a previous report.35 In addition, this study included the more recent IM method. Further, the particle size estimates of sGGE, IM, and NMR had good correlation and demonstrated good correlation of estimates of particle size with TG, HDL-C, and TG/HDL-C ratio. In summary, we found substantial agreement in the reported LDL phenotype among 4 different lipid subfraction technologies offered by 4 clinical reference laboratories.
The authors thank Colette Scheele (Quest Diagnostics Nichols Institute), Junaid Shabeer and Jean Amos Wilson (Berkeley HeartLab) and their staff for subject recruitment, phlebotomy, and sample preparation and tracking. Andre Arellano, Dov Shiffman, and Jim Devlin (Celera) provided valuable manuscript comments.
1. Anitschkow N, Chalatow S. Ueber experimentelle cholesterinsteatose und ihre bedeutung für die entstehung einiger pathologischer prozesse. Centrbl Allg Pathol Pathol Anat. 1913; 24: 1–9.
2. National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). 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.
3. Sachdeva A, Cannon CP, Deedwania PC, et al. Lipid levels in patients hospitalized with coronary artery disease: an analysis of 136,905 hospitalizations in Get with the guidelines. Am Heart J. 2009; 157: 111–117.
4. Superko HR, King S III. Lipid management to reduce cardiovascular risk: a new strategy is required. Circulation. 2008; 117: 560–568.
5. Gofman JW, Young W, Tandy R. Ischemic heart disease, atherosclerosis, and longevity. Circulation. 1966; 34: 679–697.
6. Shen MM, Krauss RM, Lindgren FT, et al. Heterogeneity of serum low density lipoproteins in normal human subjects. J Lipid Res. 1981; 22: 236–244.
7. Krauss RM, Burke DJ. Identification of multiple subclasses of plasma low density lipoproteins in normal humans. J Lipid Res. 1982; 23: 97–104.
8. Superko HR. What can we learn about dense LDL and lipoprotein particles from clinical trials? Curr Opin Lipidol. 1996; 7: 363–368.
9. Nishina PM, Johnson JP, Naggert JK. Linkage of atherogenic lipoprotein phenotype to the low-density lipoprotein receptor locus on the short arm of chromosome 19. Proc Natl Acad Sci U S A. 1992; 89: 708–712.
10. Zhao XQ, Kosinski AS, Barnhart HX, et al. Prediction of native coronary artery disease progression following PTCA or CABG in the Emory Angioplasty Versus Surgery Trial. Med Sci Monit. 2003; 9: 48–54.
11. Williams PT, Zhao XQ, Marcovina SM, et al. Levels of cholesterol in small LDL particles predict atherosclerosis progression and incident CHD in the HDL-Atherosclerosis Treatment Study (HATS). PLoS One. 2013; 8: e56782
12. Austin MA, Breslow JL, Hennekens CH, et al. Low-density lipoprotein subclass patterns and risk of myocardial infarction. JAMA. 1988; 260: 1917–1921.
13. Stampfer MJ, Krauss RM, Ma J, et al. A prospective study of TG level, low-density lipoprotein particle diameter, and risk of myocardial infarction. JAMA. 1996; 276: 882–888.
14. Gardner CD, Fortmann SP, Krauss RM. Association of small low-density lipoprotein particles with the incidence of coronary artery disease in men and women. JAMA. 1996; 276: 875–881.
15. St-Pierre AC, Cantin B, Dagenais GR, et al. Low-density lipoprotein subfractions and the long-term risk of ischemic heart disease in men 13-year follow-up data from the Québec Cardiovascular Study. Arterioscler Thromb Vasc Biol. 2005; 25: 553–559.
16. Austin MA. Triglyceride, small, dense low-density lipoprotein, and the atherogenic lipoprotein phenotype. Curr Atheroscler Rep. 2000; 2: 200–207.
17. Lamarche B, Tchernof A, Moorjani S, et al. Small, dense low-density lipoprotein particles as a predictor of the risk of ischemic heart disease in men. Prospective results from the Quebec Cardiovascular Study. Circulation. 1997; 95: 69–75.
18. Balk E, Ip S, Chung M, et al. Low density lipoprotein subfractions: systematic review of measurement methods and association with cardiovascular outcomes. Technology Assessment Report. 2008; 1–129.
19. Superko HR, Gadesam RR. Is it LDL particle size or number that correlates with risk for cardiovascular disease? Curr Atheroscler Rep. 2008; 10: 377–385.
20. Ip S, Lichtenstein AH, Chung M, et al. Systematic review: association of low-density lipoprotein subfractions with cardiovascular outcomes. Ann Intern Med. 2009; 150: 474–484.
21. Superko HR. Advanced lipoprotein testing and subfractionation are clinically useful. Circulation. 2009; 119: 2383–2395.
22. Mora S. Advanced lipoprotein testing and subfractionation is not (yet) ready for routine clinical use. Circulation. 2009; 119: 2396–2404.
23. Krauss R. Lipoprotein subfractions and cardiovascular disease risk. Curr Opin Lipidol. 2010; 21: 305–311.
24. Greenland P, Alpert JS, Beller GA, et al. 2010 ACCF/AHA guideline for assessment of cardiovascular risk in asymptomatic adults: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines. Circulation. 2010; 122: e584–e636.
25. Davidson MH, Ballantyne CM, Jacobson TA, et al. Clinical utility of inflammatory markers and advanced lipoprotein testing: advice from an expert panel of lipid specialists. J Clin Lipidol. 2011; 5: 338–367.
26. Mikhailidis DP, Elisaf M, Rizzo M, et al. “European panel on low density lipoprotein (LDL) subclasses”: a statement on the pathophysiology, atherogenicity and clinical significance of LDL subclasses. Curr Vasc Pharmacol. 2011; 9: 1–39.
27. Reiner Z, Catapano AL, De Backer G, et al. ESC/EAS Guidelines for the management of dyslipidemias. Eur Heart J. 2011; 32: 1769–1818.
28. Tsai MY, Cao J. Lipoprotein subclasses—a changing landscape. Curr Cardiovasc Risk Rep. 2008; 2: 23–28.
29. Catapano AL, Reiner Z, De Backer G, et al. ESC/EAS Guidelines for the management of dyslipidaemias. The task force for the management of dyslipidaemias of the European Society of Cardiology (ESC) and the European Atherosclerosis Society (EAS). Atherosclerosis. 2011; 217: 3–46.
30. Miller M, Stone NJ, Ballantyne C, et al. Triglycerides and cardiovascular disease: a scientific statement from the American Heart Association. Circulation. 2011; 123: 2292–2333.
31. Caulfield MP, Li S, Lee G, et al. Direct determination of lipoprotein particle sizes and concentrations by ion mobility analysis. Clin Chem. 2008; 54: 1307–1316.
32. Superko HR. Advanced lipoprotein testing and subfractionation are clinically useful. Circulation. 2009; 119: 2383–2395.
33. Hanak V, Munoz J, Teague J, et al. Accuracy of the TG to high-density lipoprotein cholesterol ratio for prediction of the low-density lipoprotein phenotype B. Am J Cardiol. 2004; 94: 219–222.
34. Boizel R, Benhamou PY, Lardy B, et al. Ratio of TGs to HDL cholesterol is an indicator of LDL particle size in patients with type 2 diabetes and normal HDL cholesterol levels. Diabetes Care. 2000; 23: 1679–1685.
35. Ensign W, Hill N, Heward CB. Disparate LDL phenotypic classification among 4 different methods assessing LDL particle characteristics. Clin Chem. 2006; 52: 1722–1727.
36. Kulkarni KR, Garber DW, Jones MK, et al. Identification and cholesterol quantification of low density lipoprotein subclasses in young adults by VAP-II methodology. J Lipid Res. 1995; 36: 2291–2302.
37. Austin MA, King MC, Vranizan KM, et al. Atherogenic lipoprotein phenotype: a proposed genetic marker for coronary heart disease risk. Circulation. 1990; 82: 495–506.
38. Chung BH, Segrest JP, Ray MJ, et al. Single vertical spin density gradient ultracentrifugation. Methods Enzymol. 1986; 128: 181–209.
39. Krauss RM, Blanche PJ. Detection and quantitation of LDL subfractions. Curr Opin Lipidol. 1992; 3: 377–383.
40. Nichols AV, Krauss RM, Musliner TA. Nondenaturing polyacrylamide gradient gel electrophoresis. Methods Enzymol. 1986; 128: 417–431.
41. Lee DM, Downs D. A quick and large-scale density gradient subfractionation method for low density lipoproteins. J Lipid Res. 1982; 23: 14–27.
42. Otvos JD, Jeyarajah EJ, Bennett DW, et al. Development of a proton nuclear magnetic resonance spectroscopic method for determining plasma lipoprotein concentrations and subspecies distributions from a single, rapid measurement. Clin Chem. 1992; 38: 1632–1638.
43. Jeyarajah EJ, Cromwell WC, Otvos JD. Lipoprotein particle analysis by nuclear magnetic resonance spectroscopy. Clin Lab Med. 2006; 26: 847
44. Cohen JA. Coefficient of agreement for nominal scales. Educ Psychol Meas. 1960; 20: 37–46.
45. Fleiss JL, Cohen J, Everitt BS. Large sample standard errors of kappa and weighted kappa. Psychol Bul. 1969; 72: 323–327.
46. Wilcox RR, Tian T. Comparing dependent correlations. J Gen Psychol. 2008; 135: 105–112.