- Email: [email protected]
Comparability of methods for LDL subfraction determination: A systematic review
Atherosclerosis 205 (2009) 342–348
Contents lists available at ScienceDirect
Atherosclerosis journal homepage: www.elsevier.com/locate/atherosclerosis
Review
Comparability of methods for LDL subfraction determination: A systematic review夽 Mei Chung a , Alice H. Lichtenstein a,b , Stanley Ip a , Joseph Lau a , Ethan M. Balk a,∗ a Tufts Evidence-based Practice Center, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, 800 Washington Street, Boston, MA 02111, United States b Jean Mayer USDA Human Nutrition Research Center on Aging, Tufts University, 711 Washington Street, Boston, MA 02111, United States
a r t i c l e
i n f o
Article history: Received 18 September 2008 Received in revised form 5 November 2008 Accepted 8 December 2008 Available online 14 December 2008 Keywords: LDL subfractions Clinical chemistry tests Systematic review
a b s t r a c t Identifying and aggressively treating individuals at elevated risk of developing cardiovascular disease (CVD) is critical to optimizing health outcomes. The CVD risk factors defined by the National Cholesterol Education Program do not fully predict individuals at high risk of developing CVD. Validation of potential methodologies against a reference method is essential to the adoption of a potential new risk factor to improve risk prediction. Low-density lipoprotein (LDL) subfraction has been advanced as a potential additional CVD risk factor. Currently, there is no reference method for determining LDL subfractions or standardizing the different methods used to measure LDL subfractions. We conducted a systematic review to identify reports comparing two or more methods of measuring LDL subfractions. Nine articles were identified that separated and quantified LDL subfractions by at least two methods. Comparative data were available for nuclear magnetic resonance vs. gel electrophoresis (GE), LipoPrint® vs. other GE methods, ultracentrifugation vs. GE, and high performance gel filtration chromatography vs. GE. We found a wide range of agreement (from 7 to 94% concordance for classifying LDL patterns) among methods for LDL subfraction determinations. Different criteria and definitions were used among the articles to classify individuals with respect to CVD risk. No study used CVD or other clinical outcomes as an outcome measure. In summary, the currently available literature does not provide adequate data about comparability in terms of test performance to choose one or another method to serve as a standard nor are data on comparability in terms of predicting CVD outcomes. © 2009 Elsevier Ireland Ltd. All rights reserved.
Contents 1. 2.
3.
Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Statistical issues inherent to the eligible studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Nuclear magnetic resonance (NMR) vs. gel electrophoresis (GE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. LipoPrint® GE vs. other GE methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4. Ultracentrifugation vs. GE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5. High performance gel filtration chromatography (HPLC) vs. GE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conflict of interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix A. Supplementary data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
343 343 343 346 346 347 347 347 348 348 348 348
夽 The authors of this report are responsible for its content. Statements in the report should not be construed as endorsement by the Agency for Healthcare Research and Quality or the U.S. Department of Health and Human Services. ∗ Corresponding author at: Tufts Evidence-based Practice Center, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, Box 63, 800 Washington Street, Boston, MA 02111, United States. E-mail address: [email protected] (E.M. Balk). 0021-9150/$ – see front matter © 2009 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.atherosclerosis.2008.12.011
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Measurements of low-density lipoprotein (LDL) subfractions may provide additional predictive power to LDL cholesterol concentration measurement alone or in combination with other risk factors to estimate cardiovascular disease (CVD) risk [1]. To date there is no consensus on this issue. In addition, there is no reference method to adjudicate the potential value of LDL subfraction measurements as an additional risk factor to those currently defined by the National Cholesterol Education Program [2]. Small dense LDL (sdLDL) particles have been reported to confer a higher level of CVD risk than larger less dense LDL particles [3,4]. In vitro, sdLDL particles are more avidly taken up by macrophages than larger less dense LDL particles [5]. Furthermore, sdLDL particles have been reported to be more susceptible to oxidative modification, have a greater propensity for transport into the subendothelial space, and have a greater binding potential to arterial wall proteoglycans than the larger less dense LDL particles [6]. Thus, sdLDL particles are thought to be more atherogenic than the larger less dense LDL particles. Multiple terms are used to describe LDL subfraction distributions, subparticles and characteristics, including LDL subclasses, particle concentration, particle number, particle diameter, particle density, and patterns or phenotypes. These terms describe separate, but often overlapping features of the LDL particle. For simplicity, this report uses the generic term subfractions, except where specific measurements are being described. A variety of methods, including gel electrophoresis and ultracentrifugation, are being used to measure LDL subfraction distributions, density, concentrations or diameter. LDL patterns or phenotypes can be categorized based on LDL size cutoffs, subfraction distributions, or algorithms such as electrophoretic mobility values. In December 2006, the Food and Drug Administration (FDA) held a public hearing on lipoprotein subfractions (www.fda.gov/ OHRMS/DOCKETS/ac/06/transcripts/2006-4263t1-01t.pdf, accessed July 16, 2008). After that meeting, the Centers for Medicare & Medicaid Services (CMS) requested a systematic review of the literature on LDL subfractions and the risk of CVD [1]. This article addresses the issue of test performance comparisons of different methods for measuring LDL subfractions. 1. Methods We conducted a comprehensive search of the scientific literature to identify relevant studies of test performance for measuring LDL subfractions in Medline (from 1950), CAB Abstracts (from 1973), and the Cochrane Clinical Trial Registry (3rd quarter 2007). Search terms for LDL, particle size or subfractions, and test methodologies are included in the Supplementary Material. The literature searches were limited to humans and English language publications. The final search was performed in June 2008. We included any method designated as measuring LDL subfraction distribution that used human serum or plasma samples. For the purposes of our analyses we divided these methods into different general categories (Table 1). To qualify for inclusion in this analysis, eligible studies had to compare methods from two or more different categories of methods. Because of the clinical availability of the methods, for the purpose of our analyses, LipoPrintTM and Berkeley HeartLab® gel electrophoresis (GE) were considered different categories of methods than “bench” GE (which are used exclusively in research laboratories). In addition, studies had to use serum samples from at least 10 adults for each method. We excluded studies that evaluated only incremental or technical changes to the methods (e.g., comparison of LDL particle size by HPLC with ultraviolet light detection to a modified method based on selective detection of lipoproteins by postcolumn labeling with a fluorescent lipid probe). All statistical methods of comparing test performance were evaluated, acknowledging that different methods of comparing test
343
performance make different statistical assumptions and this may have caused different interpretations of the results. We do not describe in detail the laboratory methods used. This type of detailed information is contained in the original articles. 2. Results The literature searches yielded 6615 citations, of which 472 were retrieved for further consideration for this and other related research questions in the original report to CMS [1]. Of these, nine articles provided data on the comparison of different methods [7,9,11–16,18]. All studies measured correlation or compared agreement (variously determined) between LDL subfraction measurement methods or technical efficacy [19]. None compared the diagnostic accuracy efficacy; for example, how often the tests would correctly diagnose a clinical condition such as CVD. Table 2 summarizes the nine articles that compared different methods for measuring LDL subfractions. LDL subfractions were quantified on either continuous (e.g., LDL size) or categorical (e.g., LDL patterns) scales. 2.1. Statistical issues inherent to the eligible studies To help interpret the study results, we first describe the statistical tests that were used by the eligible studies. Six of the nine articles reported correlation coefficients between a test method and a “reference standard” to determine whether the methods of measurement can be used interchangeably [7,11,14–17]. Notably, correlation is a measure of the strength of a relation (or association) rather than the agreement between two tests. Due to the underlying statistical assumptions for correlation coefficients, interpretation of test accuracy using these values can be misleading because a correlation coefficient depends on the range of values in the sample (e.g., range of LDL subfraction distribution) [20]. If the range is wide, the correlation is likely to be greater than if it is narrow. Furthermore, assigning a correlation coefficient threshold value for widespread acceptance of a test is difficult. In five of the nine studies, the percent concordance of classifying individual subjects’ LDL pattern (e.g., pattern B, AB, or A) between two tests were reported [9,13–15,18]. Zero percent concordance indicates that none of the individual subjects were classified as having the same LDL pattern by the two tests, while 100% concordance indicates the two tests classified the same LDL pattern for all individual subjects. This simple calculation of agreement does not account for either the inter- or intra-test variability. Other measures of agreement that account for test precisions, such as Bland–Altman limits of agreement (B–A LOA) [20] and the Kappa statistic [21], were reported in only three articles. These measures provide more complete data on test agreements than correlation or concordance, especially when there is no reference standard. B–A LOA, used by two studies [7,12], consist of the calculation of both the mean difference between each pair of measurements (the bias), and the standard deviation of the differences (the precision). The LOA indicate the range in which 95% of the differences between the two tests can be expected to fall. Zero bias indicates that on average across the range of results, the two tests agree. An important interpretation of B–A LOA graphs is whether the bias (disagreement between the two tests) varies across the range of results. Ideal tests would have consistent bias across wide range of values. The kappa statistic, used by one study [9], is an index of agreement between two tests taking into account any agreement that may occur by chance (or inter- and intra-test variability). It can be thought of as the chance-corrected proportional agreement, and possible values range from +1 (perfect agreement) via 0 (no agreement above that expected by chance) to −1 (complete disagreement).
344
Table 1 Methods of measuring LDL subfractions in published literature. Berkeley HeartLab® gradient gel electrophoresis This is a standardized system using a specific gradient gel electrophoresis to provide LDL subfraction patterns. The standardized version of this system is performed only at the Berkeley HeartLab® , but is clinically available.
High pressure liquid chromatography (HPLC) The original HPLC method for measuring LDL size monitors the column effluent at 280 nm of the isolated LDL subfraction by ultracentrifugation [7]. A drawback of this method is the necessity of LDL isolation by ultracentrifugation prior to chromatography. A modified HPLC method that is based on selective detection of lipoproteins by postcolumn labeling with parinaric acid (a fluorescent lipid probe) permits measurement of LDL size in whole plasma or serum [8]. Notably, though, the method has only rarely been used over the past decade by researchers of LDL subfractions and risks of cardiovascular diseases. LipoPrint® This is a clinically available measurement technique that uses a standardized method for using linear polyacrylamide gel electrophoresis to separate LDL particles on the basis of size and to a lesser extent charge. The kit and instrument for this method are marketed by Quantimetrix. This method permits separation of LDL into seven subfractions within 60 min. Multiple samples can be run simultaneously. Because the gels are prepared by the company, it is technically simpler, less resource intensive, and more conducive to routine laboratory testing than traditional GE [9]. Quantimetrix provides a standardized reporting system, but research laboratories commonly modify these. Nuclear magnetic resonance (NMR) This method is available for clinical use in a small number of medical laboratories. NMR measures the signal from the aggregate number of terminal methyl groups in the lipid within the particle. The number of methyl groups is reflected in the amplitude of the methyl NMR signal. The amplitude of each lipoprotein particle signal serves as a measure of the concentration of that lipoprotein. Using standard assumptions concerning lipoprotein diameter and lipid content, the NMR data can be transformed (through proprietary calculations) into subfraction concentrations. Other quantitative subfraction information, such as LDL size/diameter and patterns, can also be derived through additional calculations [10]. There are, however, many layers of assumptions within the NMR software, which is proprietary. Some of the unknown assumptions, including calibration and validation issues, have been addressed [10] but some remain to be fully evaluated. Ultracentrifugation This covers a wide range of methods that separate lipoprotein particles on the basis of density, either sequentially and continuously, prior to lipid or apoprotein analysis. These methods are time- and resource-intensive. Other techniques Examples include Vertical Auto Profile® , LipophorTM (another gel electrophoresis method developed by Quantimetrix), and capillary isotachophoresis (CITP).
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Gel electrophoresis (“bench”) This covers a wide range of methods using gel electrophoresis. These methods are either not standardized or, if standardized, are not routinely used by clinical laboratories. In general, researchers prepare their own gels and use their own methods for running the analyses. Different compounds are used to create the gels, though polyacrylamide is most common, and different distributions of gel densities are used. These methods tend to be time- and resource-intensive.
Table 2 Comparison of different methods for measuring LDL subfractions. Author, country
N
Mean (range), mg/dL LDL-c
NMR vs. GE Blake et al. [11], US Witte et al. [12], Netherlands
Ensign et al. [13], US
Davy and Davy [15], US
LipoPrint® vs. other GE Hirany et al. [9], US
Hoefner et al. [14], US
Ensign et al. [13], US Ultracentrifugation vs. GE Dormans et al. [16], Netherlands Ensign et al. [13], US
O’Neal et al. [17], Australia
Population Tg
21
111
117
164
324
nd
nd
≤545
37–40
51
113
102
51
(58–820)
120
94
125 (42–452)
120
nd
213
156
219 (113–563)
213
(37–479)
217
100
270 (61–617)
217
Tests Test 1 (metric)
Convenience sample of healthy people Case: diabetes (type 1) Control: a random sample from general population
Convenience sample of healthy people
nd
Men (62%) and women in a cross-sectional research study: age 18–70 years old; BMI 17.8–39.9; 15% metabolic syndrome
nd
nd
Subgroup or subanalysis
Analysis
Results
Test 2 or “Ref Std” (metric)
NMR (size, nm)
GE (bench) (size, nm)
All
Correlation, r
0.89 (P < .001)
NMR (size, nm)
GE (bench) (size, nm)
All
B–A LOA (95% CI)
−5.38 (−6.79, −3.97)
Type 1 DM (n = 152) No DM (n = 172) Men (n = 156) Women (n = 168) Tg < 79 mg/dL (n = 108) Tg 79–118 (n = 109) Tg > 118 (n = 107)
B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI)
−5.49 (−7.31, −3.68) −5.27 (−6.96, −3.60) −5.20 (−6.86, −3.53) −5.55 (−7.41, −3.68) −5.73 (−7.54, −3.92) −5.41 (−6.94, −3.89) −4.99 (−6.61, −3.37)
All
Concordance
70% (28/40)
GE pattern A vs. B
Concordance
51% (19/37)
GE pattern A or AB vs. B GE pattern A vs. AB or B
Concordance Concordance
80% (32/40) 54% (20/37)
All
Correlation, r
−0.67 (P < .001)
Pattern A Intermediate Pattern B
Concordance Concordance Concordance
94% 7% 67%
All
Correlation, r
0.92 (P < .01)f
NMR vs. GE
Categorized individuals as having the pattern B phenotype
29% vs. 15%, P < .0001
All
Weighted kappa (95% CI)
0.78 (0.68, 0.87)
Small Intermediate Large
Concordance Concordance Concordance
92% 33% 77%
Pattern A
Concordance
88%
Intermediate Pattern B
Concordance Concordance
64% 24%
NMR (pattern A or B)
NMR (pattern A, intermediate, pattern B based on absolute size cutoffs)c
NMR (pattern A, B based on LDL size cutoffs)d
LipoPrint® GE (small, intermediate; large based on Rf cutoff values)g
LipoPrint® GE (pattern A, intermediate, pattern B based on LDLSF score)b
GE (bench) (pattern A, AB, or B)a LipoPrintTM (pattern A, AB, or B)b
LipoPrintTM (pattern A, intermediate, pattern B based on LDLSF score)b
Berkeley HeartLab GE (pattern A, AB, B based on LDL size cutoffs)e
GE (bench) (small, intermediate; large based on absolute size cutoffs)h
GE-ZaxisTM (pattern A & B per Berkley HeartLab cutpoints)
(58–820)
nd
(37–479)
Convenience sample of healthy people
LipoPrint® GE (pattern A, AB, or B)b
GE (bench) (pattern A, AB, or B)a
All
Concordance
40% (14/35)
41
nd
213
143
0.85 (P < .001)
nd
(37–479)
GE (bench) (migration distance, mm) GE (bench) (pattern A, AB, or B)a LipoPrintTM (pattern A, AB, or B)b GE (bench) (size, nm)
Correlation, r
(58–820)
DGUC (LDL-1, LDL-2 or LDL-3, g/mL) UC-VAP-II (pattern A, AB, or B)i
All
37
Convenience sample of healthy people Convenience sample of healthy people
All
Concordance
41% (15/37)
All
Concordance
11% (4/37)
All
Correlation, r
0.78 (P < .0001)
Vertical DGUC vs. GE
Mean size
23.1 vs. 26.1 nm, P < 0.0001
27
nd
nd
(61–213)
Convenience sample of patients with type 2 diabetes (26%) or from the general population
Vertical DGUC with light-scattering methodology (size, nm)
345
35
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Hoefner et al. [14], US
TC
0.25 (−0.6, 1.0) B–A LOA (95% CI) All
LDL-c, low-density lipoprotein cholesterol; TC, total cholesterol; Tg, triglycerides; Ref Std, reference standard; r, correlation coefficient; B–A LOA, Bland–Altman limits of agreement [20]; kappa, kappa statistic [21]; DGUC, density-gradient ultracentrifugation; nd, no data; DM, diabetes; VAP, Vertical Auto Profile® ; GE (bench), a research laboratory-specific method of gel electrophoresis (not clinically available); AUC, area under the curve; HPLC, high performance gel filtration chromatography; NMR, nuclear magnetic resonance; GE, gel electrophoresis. a Large LDL (pattern A): 26.35–28.5 nm; intermediate LDL (pattern AB): 25.75–26.34 nm; small LDL (pattern B): 22.0–25.74 nm. b Pattern A: LDLSF score <5.5; intermediate: 5.5–8.5; pattern B: >8.5. c Pattern A: 20.6–22.0 nm; intermediate: 20.4–20.5; pattern B: 19.0–20.3 nm. d Pattern A: LDL 20.6–22.0 nm; pattern B: ≤20.5 nm. e Pattern A: LDL 26.3–28.5 nm; AB: 25.75–26.34 nm; pattern B: <22.0–25.75 nm. f Relationship of apoprotein B concentration (GE) to LDL particle number (NMR). g Small LDL: R > 0.40, intermediate LDL: R = 0.38–0.40, large LDL: R < 0.38. f f f h Small LDL: <25.8 nm, intermediate LDL: 25.8–26.3 nm, large LDL: >26.3 nm. i LDL1 (most buoyant) through LDL 6 (most dense): LDL1 and LDL2 comprise pattern A; LDL3 and LDL4 comprise pattern B. j The authors used both salt DGUC and GE as the reference standard in the calculation of the test (iodixanol DGUC) performance (“sensitivity” and “specificity”). The interpretations of the “sensitivity” and “specificity” are same as concordance when no true reference standard.
92%
0.88 (P < .001)
“Specificity”j All
HPLC (size, nm) Convenience sample of patients with diabetes (type 2) 209 (45–509) 231 (135–315) nd 60 HPLC vs. GE Scheffer et al. [7], Netherlands
Iodixanol DGUC (peak density >1.028 kg/L)
GE (bench) (size, nm)
Correlation, r
100% 94% “Specificity”j “Sensitivity”j All All
All
100% “Sensitivity”j GE (bench) or salt DGUCj (sd LDL-III, or LDL subfraction pattern B) Davies et al. [18], UK
47
nd
nd
nd
Convenience sample of healthy people
Iodixanol DGUC (>51% AUC LDL density >1.028 kg/L)
Analysis Subgroup or subanalysis Test 2 or “Ref Std” (metric) Tests
Test 1 (metric)
Population Tg TC LDL-c
Mean (range), mg/dL N Author, country
Table 2 (Continued)
All
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Results
346
2.2. Nuclear magnetic resonance (NMR) vs. gel electrophoresis (GE) Five articles reported six comparisons of NMR and GE (bench, LipoPrint® and HeartLab® ) involving 549 subjects (Table 2, NMR vs. GE) [10–14]. The lipoprotein profiles of the 549 subjects across studies were heterogeneous. The units of measure included particle sizes/diameters and patterns/phenotypes. Although a good correlation between LDL particle sizes assessed by NMR and bench GE (using a method specific to their research laboratory) was found in 21 apparently healthy men (r = 0.89, P < .001) reported by Blake et al. [11], Witte et al. found that the mean difference (or mean bias by B–A LOA analysis) between measured LDL size on NMR and peak LDL size on GE was 53.8 Å (with NMR yielding smaller measurements) in 324 men and women with and without type 1 diabetes [12]. The 95% LOA were 39.7 and 67.9 Å. The difference in LDL size according to NMR and GE and the strength of the relationships between the two were different in different study population subgroups, suggesting inconsistent agreement between NMR and GE across various populations. The mean differences were larger for patients with type 1 diabetes, women, and those with lower triglyceride concentrations. Ensign et al. and Hoefner et al. studies, with a total of 90 subjects, showed a fair to good concordance or agreement between NMR-assessed and LipoPrint® or bench GE-assessed LDL patterns (ranging from 51 to 94%) [13,14]. The study by Davy et al. of 131 subjects showed that NMR was more likely than Berkeley HeartLab® GE to categorize individuals as having the pattern B phenotype, corresponding to sdLDL (29% vs. 15%, P < .0001). The wide range of agreement found across studies may be partly explained by the heterogeneity in the classifications of LDL patterns between the different methods. NMR studies classified LDL patterns based on standard absolute size cutoffs, but some studies used two categories (patterns A and B) while others used three categories (patterns A, intermediate, and B). Studies that utilized GE to measure LDL subfractions used the same classification scheme but with different size cutoffs. Studies using LipoPrint® classified LDL patterns based on a complex LDL score derived from area under the curve at various predefined electrophoretic mobility (Rf ) values. 2.3. LipoPrint® GE vs. other GE methods Three articles compared LipoPrint® and other methods of GE LDL subfraction separation involving a total of 188 subjects (Table 2, LipoPrint® vs. other GE) [9,13,14]. All three studies evaluated the concordance or agreement between LDL patterns assessed by LipoPrint® and other GE. However, in all cases the Lipoprint® kit was not used according to the manufacturer’s instructions. The investigators used separate criteria to evaluate and classify the results of the Lipoprint® test. Hirany et al. classified LDL subfractions into small, intermediate or large based on electrophoretic mobility (Rf ) cutoffs (Table 2). Hoefner et al. and Ensign et al. classified LDL subfractions into pattern A, AB, and B based on an LDL subfraction score (LDLFS) [13,14]. The concordance rates between LipoPrint® assessed and GE-assessed LDL patterns varied according to the LDL phenotypes. Hirany et al. [9] reported good agreement between LipoPrint® and another GE method after evaluating the data using kappa statistics (weighted kappa = 0.78; 95% CI, 0.68–0.87). LipoPrint® had an agreement of 92% concordance for classification of the small LDL subfraction compared with GE. For large LDL subfraction, LipoPrint® had an agreement of 77% concordance compared with GE. Hoefner et al. [14] reported 84, 64, and 24% agreement for classification of the small, intermediate and large LDL subfraction, respectively, for LipoPrint® and GE. Ensign et al. reported a 40%
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
agreement in the classification of LDL patterns between LipoPrint® and GE. 2.4. Ultracentrifugation vs. GE Four articles reported five comparisons of ultracentrifugation and various GE methods involving a total of 152 subjects (Table 2, ultracentrifugation vs. GE) [13,16–18]. The lipid profiles of these 152 subjects across studies varied greatly although the data were incompletely reported in most studies. There was no uniform ultracentrifugation or GE methodology across studies. Therefore, the results from these five comparisons are summarized individually. Dormans et al. [16] demonstrated that migration distance of the predominant LDL subfraction from bench GE correlated strongly with the density of the predominant LDL band derived from ultracentrifugation (r = 0.85, P < .0001) in 41 healthy individuals. Ensign et al. [13] reported 41% agreement for classification of LDL patterns between ultracentrifugation vertical auto profile and bench GE, and 11% agreement for classification of LDL patterns between ultracentrifugation vertical auto profile and LipoPrint® . O’Neal et al. [17] showed a good correlation (r = 0.78, P < .0001) when comparing vertical ultracentrifugation and light-scattering methodology with bench GE for determining LDL particle size. However, the mean LDL size obtained by vertical ultracentrifugation was smaller than those obtained by GE (231 Å vs. 261 Å, P < 0.0001). Davies et al. [18] examined the diagnostic test performance of an LDL peak density of >1.025 kg/L and area under the LDL profile (>1.028 kg/L) by iodixanol density-gradient ultracentrifugation in predicting a predominance of small dense LDL III (or pattern B) as determined by bench GE or salt-density ultracentrifugation. However, this study provided an inadequate description of the reference standard. An area under the LDL profile of over 51% (density > 1.028 kg/L) was shown to give 100% specificity and sensitivity in differentiating a predominance of small dense LDL III (or pattern B). This was reported to be “marginally better” as a predictor of small dense LDL III than the cutoff density of 1.028 kg/L alone (94% sensitivity; 92% specificity). 2.5. High performance gel filtration chromatography (HPLC) vs. GE One report compared LDL particle size/diameter measured by both HPLC and bench GE in 60 patients with type 2 diabetes (Table 2, HPLC vs. GE) [7]. The total cholesterol and triglyceride concentrations ranged from 135 to 315 mg/dL and 45 to 509 mg/dL, respectively. LDL size determinations were highly correlated between the two methods (r = 0.88, P < .0001). B–A LOA showed that the mean difference between LDL size on HPLC and on GE was 2.5 Å (with HPLC yielding larger sizes). The 95% LOA were −6 and +10 Å. 3. Discussion Currently there is no method for LDL subfractions determination that is considered the reference or ‘gold’ standard. Among the methods in common use, different components or characteristics of the LDL particle are assessed. GE measures LDL subfractions by separating LDL particles on the basis of size and to a lesser extent charge. NMR measures the LDL particle number by detecting the signal from the aggregate number of terminal methyl groups of the lipid within the particle and using a proprietary algorithm to calculate size. Ultracentrifugation measures LDL subfractions by separating lipoprotein particles on the basis of density, either using sequential and continuous gradients. Comparisons among measurement
347
methods has yielded a wide range of agreement (from 7 to 94% concordance for classifying LDL patterns, mean bias [difference in pair measurements] of −5.38 Å with 95% LOA from −6.79 to −3.97 Å; weighted kappa of 0.78 with 95% confidence interval from 0.68 to 0.87) and correlation coefficients (from 0.67 to 0.92) for the comparison of NMR vs. GE and Lipoprint® vs. other GE. The differences in LOAs reported in one study [12] between methods varied across subjects with or without diabetes, making it difficult to draw conclusions from the aggregate data. The studies comparing ultracentrifugation and GE methods measured different components of the LDL particle. Hence, the wide range of agreement (concordance from 40 to 100%) and correlation coefficient (from 0.76 to 0.85) between these methods is unique to each of the individual reports and laboratories and the results do not lend themselves to easy generalizability to other laboratories using the same methods. One study that compared HPLC and GE reported good agreement between measures of LDL particle sizes [7]. On average, HPLC measurements of LDL particle size were larger than measurements based on GE with a wide LOA (from −6 to +10 Å), implying that size measurements made with the different methods are not directly comparable. The available data that evaluated comparability among and within analytical techniques are limited by the wide variety of methodologies used, non-uniform definitions or descriptions of LDL subfractions and, at times, inappropriate statistical analyses for tests of agreement. Each of the three major methods for measuring LDL subfractions in common use – GE, NMR, and ultracentrifugation – describes and measures the subfractions differently. Additionally, within a specific general type of measurement tool (e.g., GE) or within a specific method (LipoPrint® GE or NMR) there is no standardized protocol that is universally used for defining or describing the LDL subfractions. The lack of standardization among analytical methods and reference material for standardizing each analytical method may be the major contributing factors accounting for the wide range of agreements observed in the literature. Lack of harmonization and standardization precludes efforts to determine whether any biases identified are artifacts of the comparison as a result of an inadequate reference material, or are genuine biases among the methods examined [22]. Furthermore, different analytical units are used to report data on LDL subparticle distribution, including size measurements (which correlate but do not agree among methods), LDL subfraction concentrations or proportions, and patterns. In addition, research groups used different thresholds of particle diameters and number of thresholds to describe LDL subparticle distributions. It is therefore not possible to reliably compare methods with each other. Only one report directly compared all the commonly used LDL subfraction measurement methods (NMR, LipoPrint® GE, other GE, and ultracentrifugation) [13]. However, the recommended LDL subfraction definitions as defined by the manufacturer of the LipoPrint® kit were not used. Thus, generalizability to other clinical laboratories using the kit is limited. Of the nine studies included, only three [7,9,12] used appropriate statistical methods (B–A LOA or kappa statistic) to test for agreement between methods. Lastly, no study compared the diagnostic accuracy, efficacy [19], or clinical value of the LDL subparticle methods to individuals’ clinical outcomes. Therefore, the available literature does not provide any data about superiority among methods to predict CVD outcomes; nor does the literature adequately assess comparability among methods to allow for the identification of a single method for use as a ‘gold standard’. Comparisons of methods based on agreement for size or phenotypes are necessary but not sufficient to determine whether the different methods are measuring the same or similar LDL subfractions. The different combinations of physicochemical properties used to separate the lipoprotein particles by different methods (e.g., density, size/diameter, electrophoretic mobility) preclude valid
348
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
assessments of test agreement. Development of reference materials that are widely accepted as appropriate, accurate and reliable, is necessary to permit an accurate description of the similarities and differences among LDL subfraction methods. This ought to occur through a body such as National Cholesterol Standardization Program to ensure independence and credibility. After establishing a consensus reference method with standardized or harmonized analytical approaches, epidemiologic studies and clinical trials will be needed to examine the clinical value of the LDL subfraction determination to predict individuals’ clinical outcomes. Conflict of interest None. Acknowledgments Funding: This project was funded under Contract No. 290-020022 from the Agency for Healthcare Research and Quality, U.S. Department of Health and Human Services. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.atherosclerosis.2008.12.011. References [1] Balk E, Ip S, Chung M, Lau J, Lichtenstein A. Low Density Lipoprotein Subfractions: Systematic Review of Measurement Methods and Association with Cardiovascular Outcomes. Evidence Report/Technology Assessment (Prepared by Tufts Medical Center Evidence-based Practice Center, under contract No. 290-02-0022). Rockville, MD. Agency for Healthcare Research and Quality; June 16, 2008. (http://www1.cms.hhs.gov/mcd/viewtechassess.asp?from2= viewtechassess.asp&where=index&tid=56&). [2] Expert Panel on Detection Evaluation and Treatment of High Blood Cholesterol in Adults. Executive Summary of The 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). JAMA 2001;285:2486–97. [3] Krauss RM. Dense low density lipoproteins and coronary artery disease. Am J Cardiol 1995;75:53B–7B.
[4] Krauss RM. Dietary and genetic effects on low-density lipoprotein heterogeneity. Annu Rev Nutr 2001;21:283–95. [5] Chapman MJ, Guerin M, Bruckert E. Atherogenic, dense low-density lipoproteins. Pathophysiology and new therapeutic approaches. Eur Heart J 1998;19(Suppl 30). [6] Berneis KK, Krauss RM. Metabolic origins and clinical significance of LDL heterogeneity. J Lipid Res 2002;43:1363–79. [7] Scheffer PG, Bakker SJ, Heine RJ, Teerlink T. Measurement of low-density lipoprotein particle size by high-performance gel-filtration chromatography. Clin Chem 1997;43:1904–12. [8] Scheffer PG, Bakker SJ, Heine RJ, Teerlink T. Measurement of LDL particle size in whole plasma and serum by high performance gel-filtration chromatography using a fluorescent lipid probe. Clin Chem 1998;44:2148–51. [9] Hirany SV, Othman Y, Kutscher P, Rainwater DL, Jialal I, Devaraj S. Comparison of low-density lipoprotein size by polyacrylamide tube gel electrophoresis and polyacrylamide gradient gel electrophoresis. Am J Clin Pathol 2003;119:439–45. [10] Jeyarajah EJ, Cromwell WC, Otvos JD. Lipoprotein particle analysis by nuclear magnetic resonance spectroscopy. Clin Lab Med 2006;26:847–70. [11] Blake GJ, Otvos JD, Rifai N, Ridker PM. Low-density lipoprotein particle concentration and size as determined by nuclear magnetic resonance spectroscopy as predictors of cardiovascular disease in women. Circulation 2002;106:1930–7. [12] Witte DR, Taskinen MR, Perttunen-Nio H, van TA, Livingstone S, Colhoun HM. Study of agreement between LDL size as measured by nuclear magnetic resonance and gradient gel electrophoresis. J Lipid Res 2004;45:1069–76. [13] Ensign W, Hill N, Heward CB. Disparate LDL phenotypic classification among 4 different methods assessing LDL particle characteristics. Clin Chem 2006;52:1722–7. [14] Hoefner DM, Hodel SD, O’Brien JF, et al. Development of a rapid, quantitative method for LDL subfractionation with use of the Quantimetrix Lipoprint LDL System. Clin Chem 2001;47:266–74. [15] Davy BM, Davy KP. Comparison of assessment techniques: plasma lipid and lipoproteins related to the metabolic syndrome. Lipids Health Dis 2006;5:3. [16] Dormans TP, Swinkels DW, de GJ, Hendriks JC, Stalenhoef AF, Demacker PN. Single-spin density-gradient ultracentrifugation vs gradient gel electrophoresis: two methods for detecting low-density-lipoprotein heterogeneity compared. Clin Chem 1991;37:853–8. [17] O’Neal D, Harrip P, Dragicevic G, Rae D, Best JD. A comparison of LDL size determination using gradient gel electrophoresis and light-scattering methods. J Lipid Res 1998;39:2086–90. [18] Davies IG, Graham JM, Griffin BA. Rapid separation of LDL subclasses by iodixanol gradient ultracentrifugation. Clin Chem 2003;49:1865–72. [19] Fryback DG, Thornbury JR. The efficacy of diagnostic imaging. Med Decis Making 1991;11:88–94. [20] Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10. [21] Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas 1960;20:37–46. [22] Miller GW, Myers GL, Rei R. Why commutability matters. Clin Chem 2006;52: 553–4.
Contents lists available at ScienceDirect
Atherosclerosis journal homepage: www.elsevier.com/locate/atherosclerosis
Review
Comparability of methods for LDL subfraction determination: A systematic review夽 Mei Chung a , Alice H. Lichtenstein a,b , Stanley Ip a , Joseph Lau a , Ethan M. Balk a,∗ a Tufts Evidence-based Practice Center, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, 800 Washington Street, Boston, MA 02111, United States b Jean Mayer USDA Human Nutrition Research Center on Aging, Tufts University, 711 Washington Street, Boston, MA 02111, United States
a r t i c l e
i n f o
Article history: Received 18 September 2008 Received in revised form 5 November 2008 Accepted 8 December 2008 Available online 14 December 2008 Keywords: LDL subfractions Clinical chemistry tests Systematic review
a b s t r a c t Identifying and aggressively treating individuals at elevated risk of developing cardiovascular disease (CVD) is critical to optimizing health outcomes. The CVD risk factors defined by the National Cholesterol Education Program do not fully predict individuals at high risk of developing CVD. Validation of potential methodologies against a reference method is essential to the adoption of a potential new risk factor to improve risk prediction. Low-density lipoprotein (LDL) subfraction has been advanced as a potential additional CVD risk factor. Currently, there is no reference method for determining LDL subfractions or standardizing the different methods used to measure LDL subfractions. We conducted a systematic review to identify reports comparing two or more methods of measuring LDL subfractions. Nine articles were identified that separated and quantified LDL subfractions by at least two methods. Comparative data were available for nuclear magnetic resonance vs. gel electrophoresis (GE), LipoPrint® vs. other GE methods, ultracentrifugation vs. GE, and high performance gel filtration chromatography vs. GE. We found a wide range of agreement (from 7 to 94% concordance for classifying LDL patterns) among methods for LDL subfraction determinations. Different criteria and definitions were used among the articles to classify individuals with respect to CVD risk. No study used CVD or other clinical outcomes as an outcome measure. In summary, the currently available literature does not provide adequate data about comparability in terms of test performance to choose one or another method to serve as a standard nor are data on comparability in terms of predicting CVD outcomes. © 2009 Elsevier Ireland Ltd. All rights reserved.
Contents 1. 2.
3.
Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Statistical issues inherent to the eligible studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Nuclear magnetic resonance (NMR) vs. gel electrophoresis (GE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. LipoPrint® GE vs. other GE methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4. Ultracentrifugation vs. GE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5. High performance gel filtration chromatography (HPLC) vs. GE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conflict of interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix A. Supplementary data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
343 343 343 346 346 347 347 347 348 348 348 348
夽 The authors of this report are responsible for its content. Statements in the report should not be construed as endorsement by the Agency for Healthcare Research and Quality or the U.S. Department of Health and Human Services. ∗ Corresponding author at: Tufts Evidence-based Practice Center, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, Box 63, 800 Washington Street, Boston, MA 02111, United States. E-mail address: [email protected] (E.M. Balk). 0021-9150/$ – see front matter © 2009 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.atherosclerosis.2008.12.011
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Measurements of low-density lipoprotein (LDL) subfractions may provide additional predictive power to LDL cholesterol concentration measurement alone or in combination with other risk factors to estimate cardiovascular disease (CVD) risk [1]. To date there is no consensus on this issue. In addition, there is no reference method to adjudicate the potential value of LDL subfraction measurements as an additional risk factor to those currently defined by the National Cholesterol Education Program [2]. Small dense LDL (sdLDL) particles have been reported to confer a higher level of CVD risk than larger less dense LDL particles [3,4]. In vitro, sdLDL particles are more avidly taken up by macrophages than larger less dense LDL particles [5]. Furthermore, sdLDL particles have been reported to be more susceptible to oxidative modification, have a greater propensity for transport into the subendothelial space, and have a greater binding potential to arterial wall proteoglycans than the larger less dense LDL particles [6]. Thus, sdLDL particles are thought to be more atherogenic than the larger less dense LDL particles. Multiple terms are used to describe LDL subfraction distributions, subparticles and characteristics, including LDL subclasses, particle concentration, particle number, particle diameter, particle density, and patterns or phenotypes. These terms describe separate, but often overlapping features of the LDL particle. For simplicity, this report uses the generic term subfractions, except where specific measurements are being described. A variety of methods, including gel electrophoresis and ultracentrifugation, are being used to measure LDL subfraction distributions, density, concentrations or diameter. LDL patterns or phenotypes can be categorized based on LDL size cutoffs, subfraction distributions, or algorithms such as electrophoretic mobility values. In December 2006, the Food and Drug Administration (FDA) held a public hearing on lipoprotein subfractions (www.fda.gov/ OHRMS/DOCKETS/ac/06/transcripts/2006-4263t1-01t.pdf, accessed July 16, 2008). After that meeting, the Centers for Medicare & Medicaid Services (CMS) requested a systematic review of the literature on LDL subfractions and the risk of CVD [1]. This article addresses the issue of test performance comparisons of different methods for measuring LDL subfractions. 1. Methods We conducted a comprehensive search of the scientific literature to identify relevant studies of test performance for measuring LDL subfractions in Medline (from 1950), CAB Abstracts (from 1973), and the Cochrane Clinical Trial Registry (3rd quarter 2007). Search terms for LDL, particle size or subfractions, and test methodologies are included in the Supplementary Material. The literature searches were limited to humans and English language publications. The final search was performed in June 2008. We included any method designated as measuring LDL subfraction distribution that used human serum or plasma samples. For the purposes of our analyses we divided these methods into different general categories (Table 1). To qualify for inclusion in this analysis, eligible studies had to compare methods from two or more different categories of methods. Because of the clinical availability of the methods, for the purpose of our analyses, LipoPrintTM and Berkeley HeartLab® gel electrophoresis (GE) were considered different categories of methods than “bench” GE (which are used exclusively in research laboratories). In addition, studies had to use serum samples from at least 10 adults for each method. We excluded studies that evaluated only incremental or technical changes to the methods (e.g., comparison of LDL particle size by HPLC with ultraviolet light detection to a modified method based on selective detection of lipoproteins by postcolumn labeling with a fluorescent lipid probe). All statistical methods of comparing test performance were evaluated, acknowledging that different methods of comparing test
343
performance make different statistical assumptions and this may have caused different interpretations of the results. We do not describe in detail the laboratory methods used. This type of detailed information is contained in the original articles. 2. Results The literature searches yielded 6615 citations, of which 472 were retrieved for further consideration for this and other related research questions in the original report to CMS [1]. Of these, nine articles provided data on the comparison of different methods [7,9,11–16,18]. All studies measured correlation or compared agreement (variously determined) between LDL subfraction measurement methods or technical efficacy [19]. None compared the diagnostic accuracy efficacy; for example, how often the tests would correctly diagnose a clinical condition such as CVD. Table 2 summarizes the nine articles that compared different methods for measuring LDL subfractions. LDL subfractions were quantified on either continuous (e.g., LDL size) or categorical (e.g., LDL patterns) scales. 2.1. Statistical issues inherent to the eligible studies To help interpret the study results, we first describe the statistical tests that were used by the eligible studies. Six of the nine articles reported correlation coefficients between a test method and a “reference standard” to determine whether the methods of measurement can be used interchangeably [7,11,14–17]. Notably, correlation is a measure of the strength of a relation (or association) rather than the agreement between two tests. Due to the underlying statistical assumptions for correlation coefficients, interpretation of test accuracy using these values can be misleading because a correlation coefficient depends on the range of values in the sample (e.g., range of LDL subfraction distribution) [20]. If the range is wide, the correlation is likely to be greater than if it is narrow. Furthermore, assigning a correlation coefficient threshold value for widespread acceptance of a test is difficult. In five of the nine studies, the percent concordance of classifying individual subjects’ LDL pattern (e.g., pattern B, AB, or A) between two tests were reported [9,13–15,18]. Zero percent concordance indicates that none of the individual subjects were classified as having the same LDL pattern by the two tests, while 100% concordance indicates the two tests classified the same LDL pattern for all individual subjects. This simple calculation of agreement does not account for either the inter- or intra-test variability. Other measures of agreement that account for test precisions, such as Bland–Altman limits of agreement (B–A LOA) [20] and the Kappa statistic [21], were reported in only three articles. These measures provide more complete data on test agreements than correlation or concordance, especially when there is no reference standard. B–A LOA, used by two studies [7,12], consist of the calculation of both the mean difference between each pair of measurements (the bias), and the standard deviation of the differences (the precision). The LOA indicate the range in which 95% of the differences between the two tests can be expected to fall. Zero bias indicates that on average across the range of results, the two tests agree. An important interpretation of B–A LOA graphs is whether the bias (disagreement between the two tests) varies across the range of results. Ideal tests would have consistent bias across wide range of values. The kappa statistic, used by one study [9], is an index of agreement between two tests taking into account any agreement that may occur by chance (or inter- and intra-test variability). It can be thought of as the chance-corrected proportional agreement, and possible values range from +1 (perfect agreement) via 0 (no agreement above that expected by chance) to −1 (complete disagreement).
344
Table 1 Methods of measuring LDL subfractions in published literature. Berkeley HeartLab® gradient gel electrophoresis This is a standardized system using a specific gradient gel electrophoresis to provide LDL subfraction patterns. The standardized version of this system is performed only at the Berkeley HeartLab® , but is clinically available.
High pressure liquid chromatography (HPLC) The original HPLC method for measuring LDL size monitors the column effluent at 280 nm of the isolated LDL subfraction by ultracentrifugation [7]. A drawback of this method is the necessity of LDL isolation by ultracentrifugation prior to chromatography. A modified HPLC method that is based on selective detection of lipoproteins by postcolumn labeling with parinaric acid (a fluorescent lipid probe) permits measurement of LDL size in whole plasma or serum [8]. Notably, though, the method has only rarely been used over the past decade by researchers of LDL subfractions and risks of cardiovascular diseases. LipoPrint® This is a clinically available measurement technique that uses a standardized method for using linear polyacrylamide gel electrophoresis to separate LDL particles on the basis of size and to a lesser extent charge. The kit and instrument for this method are marketed by Quantimetrix. This method permits separation of LDL into seven subfractions within 60 min. Multiple samples can be run simultaneously. Because the gels are prepared by the company, it is technically simpler, less resource intensive, and more conducive to routine laboratory testing than traditional GE [9]. Quantimetrix provides a standardized reporting system, but research laboratories commonly modify these. Nuclear magnetic resonance (NMR) This method is available for clinical use in a small number of medical laboratories. NMR measures the signal from the aggregate number of terminal methyl groups in the lipid within the particle. The number of methyl groups is reflected in the amplitude of the methyl NMR signal. The amplitude of each lipoprotein particle signal serves as a measure of the concentration of that lipoprotein. Using standard assumptions concerning lipoprotein diameter and lipid content, the NMR data can be transformed (through proprietary calculations) into subfraction concentrations. Other quantitative subfraction information, such as LDL size/diameter and patterns, can also be derived through additional calculations [10]. There are, however, many layers of assumptions within the NMR software, which is proprietary. Some of the unknown assumptions, including calibration and validation issues, have been addressed [10] but some remain to be fully evaluated. Ultracentrifugation This covers a wide range of methods that separate lipoprotein particles on the basis of density, either sequentially and continuously, prior to lipid or apoprotein analysis. These methods are time- and resource-intensive. Other techniques Examples include Vertical Auto Profile® , LipophorTM (another gel electrophoresis method developed by Quantimetrix), and capillary isotachophoresis (CITP).
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Gel electrophoresis (“bench”) This covers a wide range of methods using gel electrophoresis. These methods are either not standardized or, if standardized, are not routinely used by clinical laboratories. In general, researchers prepare their own gels and use their own methods for running the analyses. Different compounds are used to create the gels, though polyacrylamide is most common, and different distributions of gel densities are used. These methods tend to be time- and resource-intensive.
Table 2 Comparison of different methods for measuring LDL subfractions. Author, country
N
Mean (range), mg/dL LDL-c
NMR vs. GE Blake et al. [11], US Witte et al. [12], Netherlands
Ensign et al. [13], US
Davy and Davy [15], US
LipoPrint® vs. other GE Hirany et al. [9], US
Hoefner et al. [14], US
Ensign et al. [13], US Ultracentrifugation vs. GE Dormans et al. [16], Netherlands Ensign et al. [13], US
O’Neal et al. [17], Australia
Population Tg
21
111
117
164
324
nd
nd
≤545
37–40
51
113
102
51
(58–820)
120
94
125 (42–452)
120
nd
213
156
219 (113–563)
213
(37–479)
217
100
270 (61–617)
217
Tests Test 1 (metric)
Convenience sample of healthy people Case: diabetes (type 1) Control: a random sample from general population
Convenience sample of healthy people
nd
Men (62%) and women in a cross-sectional research study: age 18–70 years old; BMI 17.8–39.9; 15% metabolic syndrome
nd
nd
Subgroup or subanalysis
Analysis
Results
Test 2 or “Ref Std” (metric)
NMR (size, nm)
GE (bench) (size, nm)
All
Correlation, r
0.89 (P < .001)
NMR (size, nm)
GE (bench) (size, nm)
All
B–A LOA (95% CI)
−5.38 (−6.79, −3.97)
Type 1 DM (n = 152) No DM (n = 172) Men (n = 156) Women (n = 168) Tg < 79 mg/dL (n = 108) Tg 79–118 (n = 109) Tg > 118 (n = 107)
B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI) B–A LOA (95% CI)
−5.49 (−7.31, −3.68) −5.27 (−6.96, −3.60) −5.20 (−6.86, −3.53) −5.55 (−7.41, −3.68) −5.73 (−7.54, −3.92) −5.41 (−6.94, −3.89) −4.99 (−6.61, −3.37)
All
Concordance
70% (28/40)
GE pattern A vs. B
Concordance
51% (19/37)
GE pattern A or AB vs. B GE pattern A vs. AB or B
Concordance Concordance
80% (32/40) 54% (20/37)
All
Correlation, r
−0.67 (P < .001)
Pattern A Intermediate Pattern B
Concordance Concordance Concordance
94% 7% 67%
All
Correlation, r
0.92 (P < .01)f
NMR vs. GE
Categorized individuals as having the pattern B phenotype
29% vs. 15%, P < .0001
All
Weighted kappa (95% CI)
0.78 (0.68, 0.87)
Small Intermediate Large
Concordance Concordance Concordance
92% 33% 77%
Pattern A
Concordance
88%
Intermediate Pattern B
Concordance Concordance
64% 24%
NMR (pattern A or B)
NMR (pattern A, intermediate, pattern B based on absolute size cutoffs)c
NMR (pattern A, B based on LDL size cutoffs)d
LipoPrint® GE (small, intermediate; large based on Rf cutoff values)g
LipoPrint® GE (pattern A, intermediate, pattern B based on LDLSF score)b
GE (bench) (pattern A, AB, or B)a LipoPrintTM (pattern A, AB, or B)b
LipoPrintTM (pattern A, intermediate, pattern B based on LDLSF score)b
Berkeley HeartLab GE (pattern A, AB, B based on LDL size cutoffs)e
GE (bench) (small, intermediate; large based on absolute size cutoffs)h
GE-ZaxisTM (pattern A & B per Berkley HeartLab cutpoints)
(58–820)
nd
(37–479)
Convenience sample of healthy people
LipoPrint® GE (pattern A, AB, or B)b
GE (bench) (pattern A, AB, or B)a
All
Concordance
40% (14/35)
41
nd
213
143
0.85 (P < .001)
nd
(37–479)
GE (bench) (migration distance, mm) GE (bench) (pattern A, AB, or B)a LipoPrintTM (pattern A, AB, or B)b GE (bench) (size, nm)
Correlation, r
(58–820)
DGUC (LDL-1, LDL-2 or LDL-3, g/mL) UC-VAP-II (pattern A, AB, or B)i
All
37
Convenience sample of healthy people Convenience sample of healthy people
All
Concordance
41% (15/37)
All
Concordance
11% (4/37)
All
Correlation, r
0.78 (P < .0001)
Vertical DGUC vs. GE
Mean size
23.1 vs. 26.1 nm, P < 0.0001
27
nd
nd
(61–213)
Convenience sample of patients with type 2 diabetes (26%) or from the general population
Vertical DGUC with light-scattering methodology (size, nm)
345
35
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Hoefner et al. [14], US
TC
0.25 (−0.6, 1.0) B–A LOA (95% CI) All
LDL-c, low-density lipoprotein cholesterol; TC, total cholesterol; Tg, triglycerides; Ref Std, reference standard; r, correlation coefficient; B–A LOA, Bland–Altman limits of agreement [20]; kappa, kappa statistic [21]; DGUC, density-gradient ultracentrifugation; nd, no data; DM, diabetes; VAP, Vertical Auto Profile® ; GE (bench), a research laboratory-specific method of gel electrophoresis (not clinically available); AUC, area under the curve; HPLC, high performance gel filtration chromatography; NMR, nuclear magnetic resonance; GE, gel electrophoresis. a Large LDL (pattern A): 26.35–28.5 nm; intermediate LDL (pattern AB): 25.75–26.34 nm; small LDL (pattern B): 22.0–25.74 nm. b Pattern A: LDLSF score <5.5; intermediate: 5.5–8.5; pattern B: >8.5. c Pattern A: 20.6–22.0 nm; intermediate: 20.4–20.5; pattern B: 19.0–20.3 nm. d Pattern A: LDL 20.6–22.0 nm; pattern B: ≤20.5 nm. e Pattern A: LDL 26.3–28.5 nm; AB: 25.75–26.34 nm; pattern B: <22.0–25.75 nm. f Relationship of apoprotein B concentration (GE) to LDL particle number (NMR). g Small LDL: R > 0.40, intermediate LDL: R = 0.38–0.40, large LDL: R < 0.38. f f f h Small LDL: <25.8 nm, intermediate LDL: 25.8–26.3 nm, large LDL: >26.3 nm. i LDL1 (most buoyant) through LDL 6 (most dense): LDL1 and LDL2 comprise pattern A; LDL3 and LDL4 comprise pattern B. j The authors used both salt DGUC and GE as the reference standard in the calculation of the test (iodixanol DGUC) performance (“sensitivity” and “specificity”). The interpretations of the “sensitivity” and “specificity” are same as concordance when no true reference standard.
92%
0.88 (P < .001)
“Specificity”j All
HPLC (size, nm) Convenience sample of patients with diabetes (type 2) 209 (45–509) 231 (135–315) nd 60 HPLC vs. GE Scheffer et al. [7], Netherlands
Iodixanol DGUC (peak density >1.028 kg/L)
GE (bench) (size, nm)
Correlation, r
100% 94% “Specificity”j “Sensitivity”j All All
All
100% “Sensitivity”j GE (bench) or salt DGUCj (sd LDL-III, or LDL subfraction pattern B) Davies et al. [18], UK
47
nd
nd
nd
Convenience sample of healthy people
Iodixanol DGUC (>51% AUC LDL density >1.028 kg/L)
Analysis Subgroup or subanalysis Test 2 or “Ref Std” (metric) Tests
Test 1 (metric)
Population Tg TC LDL-c
Mean (range), mg/dL N Author, country
Table 2 (Continued)
All
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
Results
346
2.2. Nuclear magnetic resonance (NMR) vs. gel electrophoresis (GE) Five articles reported six comparisons of NMR and GE (bench, LipoPrint® and HeartLab® ) involving 549 subjects (Table 2, NMR vs. GE) [10–14]. The lipoprotein profiles of the 549 subjects across studies were heterogeneous. The units of measure included particle sizes/diameters and patterns/phenotypes. Although a good correlation between LDL particle sizes assessed by NMR and bench GE (using a method specific to their research laboratory) was found in 21 apparently healthy men (r = 0.89, P < .001) reported by Blake et al. [11], Witte et al. found that the mean difference (or mean bias by B–A LOA analysis) between measured LDL size on NMR and peak LDL size on GE was 53.8 Å (with NMR yielding smaller measurements) in 324 men and women with and without type 1 diabetes [12]. The 95% LOA were 39.7 and 67.9 Å. The difference in LDL size according to NMR and GE and the strength of the relationships between the two were different in different study population subgroups, suggesting inconsistent agreement between NMR and GE across various populations. The mean differences were larger for patients with type 1 diabetes, women, and those with lower triglyceride concentrations. Ensign et al. and Hoefner et al. studies, with a total of 90 subjects, showed a fair to good concordance or agreement between NMR-assessed and LipoPrint® or bench GE-assessed LDL patterns (ranging from 51 to 94%) [13,14]. The study by Davy et al. of 131 subjects showed that NMR was more likely than Berkeley HeartLab® GE to categorize individuals as having the pattern B phenotype, corresponding to sdLDL (29% vs. 15%, P < .0001). The wide range of agreement found across studies may be partly explained by the heterogeneity in the classifications of LDL patterns between the different methods. NMR studies classified LDL patterns based on standard absolute size cutoffs, but some studies used two categories (patterns A and B) while others used three categories (patterns A, intermediate, and B). Studies that utilized GE to measure LDL subfractions used the same classification scheme but with different size cutoffs. Studies using LipoPrint® classified LDL patterns based on a complex LDL score derived from area under the curve at various predefined electrophoretic mobility (Rf ) values. 2.3. LipoPrint® GE vs. other GE methods Three articles compared LipoPrint® and other methods of GE LDL subfraction separation involving a total of 188 subjects (Table 2, LipoPrint® vs. other GE) [9,13,14]. All three studies evaluated the concordance or agreement between LDL patterns assessed by LipoPrint® and other GE. However, in all cases the Lipoprint® kit was not used according to the manufacturer’s instructions. The investigators used separate criteria to evaluate and classify the results of the Lipoprint® test. Hirany et al. classified LDL subfractions into small, intermediate or large based on electrophoretic mobility (Rf ) cutoffs (Table 2). Hoefner et al. and Ensign et al. classified LDL subfractions into pattern A, AB, and B based on an LDL subfraction score (LDLFS) [13,14]. The concordance rates between LipoPrint® assessed and GE-assessed LDL patterns varied according to the LDL phenotypes. Hirany et al. [9] reported good agreement between LipoPrint® and another GE method after evaluating the data using kappa statistics (weighted kappa = 0.78; 95% CI, 0.68–0.87). LipoPrint® had an agreement of 92% concordance for classification of the small LDL subfraction compared with GE. For large LDL subfraction, LipoPrint® had an agreement of 77% concordance compared with GE. Hoefner et al. [14] reported 84, 64, and 24% agreement for classification of the small, intermediate and large LDL subfraction, respectively, for LipoPrint® and GE. Ensign et al. reported a 40%
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
agreement in the classification of LDL patterns between LipoPrint® and GE. 2.4. Ultracentrifugation vs. GE Four articles reported five comparisons of ultracentrifugation and various GE methods involving a total of 152 subjects (Table 2, ultracentrifugation vs. GE) [13,16–18]. The lipid profiles of these 152 subjects across studies varied greatly although the data were incompletely reported in most studies. There was no uniform ultracentrifugation or GE methodology across studies. Therefore, the results from these five comparisons are summarized individually. Dormans et al. [16] demonstrated that migration distance of the predominant LDL subfraction from bench GE correlated strongly with the density of the predominant LDL band derived from ultracentrifugation (r = 0.85, P < .0001) in 41 healthy individuals. Ensign et al. [13] reported 41% agreement for classification of LDL patterns between ultracentrifugation vertical auto profile and bench GE, and 11% agreement for classification of LDL patterns between ultracentrifugation vertical auto profile and LipoPrint® . O’Neal et al. [17] showed a good correlation (r = 0.78, P < .0001) when comparing vertical ultracentrifugation and light-scattering methodology with bench GE for determining LDL particle size. However, the mean LDL size obtained by vertical ultracentrifugation was smaller than those obtained by GE (231 Å vs. 261 Å, P < 0.0001). Davies et al. [18] examined the diagnostic test performance of an LDL peak density of >1.025 kg/L and area under the LDL profile (>1.028 kg/L) by iodixanol density-gradient ultracentrifugation in predicting a predominance of small dense LDL III (or pattern B) as determined by bench GE or salt-density ultracentrifugation. However, this study provided an inadequate description of the reference standard. An area under the LDL profile of over 51% (density > 1.028 kg/L) was shown to give 100% specificity and sensitivity in differentiating a predominance of small dense LDL III (or pattern B). This was reported to be “marginally better” as a predictor of small dense LDL III than the cutoff density of 1.028 kg/L alone (94% sensitivity; 92% specificity). 2.5. High performance gel filtration chromatography (HPLC) vs. GE One report compared LDL particle size/diameter measured by both HPLC and bench GE in 60 patients with type 2 diabetes (Table 2, HPLC vs. GE) [7]. The total cholesterol and triglyceride concentrations ranged from 135 to 315 mg/dL and 45 to 509 mg/dL, respectively. LDL size determinations were highly correlated between the two methods (r = 0.88, P < .0001). B–A LOA showed that the mean difference between LDL size on HPLC and on GE was 2.5 Å (with HPLC yielding larger sizes). The 95% LOA were −6 and +10 Å. 3. Discussion Currently there is no method for LDL subfractions determination that is considered the reference or ‘gold’ standard. Among the methods in common use, different components or characteristics of the LDL particle are assessed. GE measures LDL subfractions by separating LDL particles on the basis of size and to a lesser extent charge. NMR measures the LDL particle number by detecting the signal from the aggregate number of terminal methyl groups of the lipid within the particle and using a proprietary algorithm to calculate size. Ultracentrifugation measures LDL subfractions by separating lipoprotein particles on the basis of density, either using sequential and continuous gradients. Comparisons among measurement
347
methods has yielded a wide range of agreement (from 7 to 94% concordance for classifying LDL patterns, mean bias [difference in pair measurements] of −5.38 Å with 95% LOA from −6.79 to −3.97 Å; weighted kappa of 0.78 with 95% confidence interval from 0.68 to 0.87) and correlation coefficients (from 0.67 to 0.92) for the comparison of NMR vs. GE and Lipoprint® vs. other GE. The differences in LOAs reported in one study [12] between methods varied across subjects with or without diabetes, making it difficult to draw conclusions from the aggregate data. The studies comparing ultracentrifugation and GE methods measured different components of the LDL particle. Hence, the wide range of agreement (concordance from 40 to 100%) and correlation coefficient (from 0.76 to 0.85) between these methods is unique to each of the individual reports and laboratories and the results do not lend themselves to easy generalizability to other laboratories using the same methods. One study that compared HPLC and GE reported good agreement between measures of LDL particle sizes [7]. On average, HPLC measurements of LDL particle size were larger than measurements based on GE with a wide LOA (from −6 to +10 Å), implying that size measurements made with the different methods are not directly comparable. The available data that evaluated comparability among and within analytical techniques are limited by the wide variety of methodologies used, non-uniform definitions or descriptions of LDL subfractions and, at times, inappropriate statistical analyses for tests of agreement. Each of the three major methods for measuring LDL subfractions in common use – GE, NMR, and ultracentrifugation – describes and measures the subfractions differently. Additionally, within a specific general type of measurement tool (e.g., GE) or within a specific method (LipoPrint® GE or NMR) there is no standardized protocol that is universally used for defining or describing the LDL subfractions. The lack of standardization among analytical methods and reference material for standardizing each analytical method may be the major contributing factors accounting for the wide range of agreements observed in the literature. Lack of harmonization and standardization precludes efforts to determine whether any biases identified are artifacts of the comparison as a result of an inadequate reference material, or are genuine biases among the methods examined [22]. Furthermore, different analytical units are used to report data on LDL subparticle distribution, including size measurements (which correlate but do not agree among methods), LDL subfraction concentrations or proportions, and patterns. In addition, research groups used different thresholds of particle diameters and number of thresholds to describe LDL subparticle distributions. It is therefore not possible to reliably compare methods with each other. Only one report directly compared all the commonly used LDL subfraction measurement methods (NMR, LipoPrint® GE, other GE, and ultracentrifugation) [13]. However, the recommended LDL subfraction definitions as defined by the manufacturer of the LipoPrint® kit were not used. Thus, generalizability to other clinical laboratories using the kit is limited. Of the nine studies included, only three [7,9,12] used appropriate statistical methods (B–A LOA or kappa statistic) to test for agreement between methods. Lastly, no study compared the diagnostic accuracy, efficacy [19], or clinical value of the LDL subparticle methods to individuals’ clinical outcomes. Therefore, the available literature does not provide any data about superiority among methods to predict CVD outcomes; nor does the literature adequately assess comparability among methods to allow for the identification of a single method for use as a ‘gold standard’. Comparisons of methods based on agreement for size or phenotypes are necessary but not sufficient to determine whether the different methods are measuring the same or similar LDL subfractions. The different combinations of physicochemical properties used to separate the lipoprotein particles by different methods (e.g., density, size/diameter, electrophoretic mobility) preclude valid
348
M. Chung et al. / Atherosclerosis 205 (2009) 342–348
assessments of test agreement. Development of reference materials that are widely accepted as appropriate, accurate and reliable, is necessary to permit an accurate description of the similarities and differences among LDL subfraction methods. This ought to occur through a body such as National Cholesterol Standardization Program to ensure independence and credibility. After establishing a consensus reference method with standardized or harmonized analytical approaches, epidemiologic studies and clinical trials will be needed to examine the clinical value of the LDL subfraction determination to predict individuals’ clinical outcomes. Conflict of interest None. Acknowledgments Funding: This project was funded under Contract No. 290-020022 from the Agency for Healthcare Research and Quality, U.S. Department of Health and Human Services. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.atherosclerosis.2008.12.011. References [1] Balk E, Ip S, Chung M, Lau J, Lichtenstein A. Low Density Lipoprotein Subfractions: Systematic Review of Measurement Methods and Association with Cardiovascular Outcomes. Evidence Report/Technology Assessment (Prepared by Tufts Medical Center Evidence-based Practice Center, under contract No. 290-02-0022). Rockville, MD. Agency for Healthcare Research and Quality; June 16, 2008. (http://www1.cms.hhs.gov/mcd/viewtechassess.asp?from2= viewtechassess.asp&where=index&tid=56&). [2] Expert Panel on Detection Evaluation and Treatment of High Blood Cholesterol in Adults. Executive Summary of The 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). JAMA 2001;285:2486–97. [3] Krauss RM. Dense low density lipoproteins and coronary artery disease. Am J Cardiol 1995;75:53B–7B.
[4] Krauss RM. Dietary and genetic effects on low-density lipoprotein heterogeneity. Annu Rev Nutr 2001;21:283–95. [5] Chapman MJ, Guerin M, Bruckert E. Atherogenic, dense low-density lipoproteins. Pathophysiology and new therapeutic approaches. Eur Heart J 1998;19(Suppl 30). [6] Berneis KK, Krauss RM. Metabolic origins and clinical significance of LDL heterogeneity. J Lipid Res 2002;43:1363–79. [7] Scheffer PG, Bakker SJ, Heine RJ, Teerlink T. Measurement of low-density lipoprotein particle size by high-performance gel-filtration chromatography. Clin Chem 1997;43:1904–12. [8] Scheffer PG, Bakker SJ, Heine RJ, Teerlink T. Measurement of LDL particle size in whole plasma and serum by high performance gel-filtration chromatography using a fluorescent lipid probe. Clin Chem 1998;44:2148–51. [9] Hirany SV, Othman Y, Kutscher P, Rainwater DL, Jialal I, Devaraj S. Comparison of low-density lipoprotein size by polyacrylamide tube gel electrophoresis and polyacrylamide gradient gel electrophoresis. Am J Clin Pathol 2003;119:439–45. [10] Jeyarajah EJ, Cromwell WC, Otvos JD. Lipoprotein particle analysis by nuclear magnetic resonance spectroscopy. Clin Lab Med 2006;26:847–70. [11] Blake GJ, Otvos JD, Rifai N, Ridker PM. Low-density lipoprotein particle concentration and size as determined by nuclear magnetic resonance spectroscopy as predictors of cardiovascular disease in women. Circulation 2002;106:1930–7. [12] Witte DR, Taskinen MR, Perttunen-Nio H, van TA, Livingstone S, Colhoun HM. Study of agreement between LDL size as measured by nuclear magnetic resonance and gradient gel electrophoresis. J Lipid Res 2004;45:1069–76. [13] Ensign W, Hill N, Heward CB. Disparate LDL phenotypic classification among 4 different methods assessing LDL particle characteristics. Clin Chem 2006;52:1722–7. [14] Hoefner DM, Hodel SD, O’Brien JF, et al. Development of a rapid, quantitative method for LDL subfractionation with use of the Quantimetrix Lipoprint LDL System. Clin Chem 2001;47:266–74. [15] Davy BM, Davy KP. Comparison of assessment techniques: plasma lipid and lipoproteins related to the metabolic syndrome. Lipids Health Dis 2006;5:3. [16] Dormans TP, Swinkels DW, de GJ, Hendriks JC, Stalenhoef AF, Demacker PN. Single-spin density-gradient ultracentrifugation vs gradient gel electrophoresis: two methods for detecting low-density-lipoprotein heterogeneity compared. Clin Chem 1991;37:853–8. [17] O’Neal D, Harrip P, Dragicevic G, Rae D, Best JD. A comparison of LDL size determination using gradient gel electrophoresis and light-scattering methods. J Lipid Res 1998;39:2086–90. [18] Davies IG, Graham JM, Griffin BA. Rapid separation of LDL subclasses by iodixanol gradient ultracentrifugation. Clin Chem 2003;49:1865–72. [19] Fryback DG, Thornbury JR. The efficacy of diagnostic imaging. Med Decis Making 1991;11:88–94. [20] Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10. [21] Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas 1960;20:37–46. [22] Miller GW, Myers GL, Rei R. Why commutability matters. Clin Chem 2006;52: 553–4.