Large HDL2 combined with inflammatory factors as superior predictors for coronary artery disease than small HDL3.

BACKGROUND: This study investigated whether combinations of high-density lipoprotein (HDL) subfractions and inflammatory markers would add value to coronary artery disease (CAD) prediction.
METHODS: Non-CAD subjects (n=245) were stratified into low/moderate/high-Framingham risk (L/M/H-FR) groups and 180 CAD patients were enrolled. Levels of HDL-C, HDL2, HDL3, monocyte chemoattractant protein-1 (MCP-1), and high-sensitivity C-reactive protein (hsCRP) were measured. Multivariable logistic models for CAD were estimated with a single parameter or all parameters together after adjustment for conventional risk factors (CRFs), and Z statistics, net reclassification improvement (NRI), and integrated discrimination improvement (IDI) were used to compare discrimination among different models.
RESULTS: The results show that HDL-C, HDL2, and HDL3 gradually decreased, while MCP-1 and hsCRP gradually increased from L/M/H-FR to the CAD group. When applying a single factor in the CRFs-adjusted models, HDL-C (OR 0.011, 95% CI, 0.002-0.071, P<0.05) and HDL2 (OR 0.000072, 95% CI, 0.000001-0.004, P<0.05), but not HDL3, were significantly related to CAD risk. Only HDL2 (OR 0.000072, 95% CI, 0.000001-0.004, P<0.001) remained significant when applying all HDL parameters. In the model including all HDL and inflammatory parameters, HDL2 (OR 0.001, 95% CI, 0.000027-0.051), MCP-1 (OR 1.066, 95% CI, 1.039-1.094), and hsCRP (OR 1.130, 95% CI, 1.041-1.227) showed significant differences (all P<0.05). This combined model showed improved discrimination over the models with a single factor (P<0.05) or all HDL parameters (Z=3.299, NRI =0.179, IDI =0.081, P<0.001).
CONCLUSIONS: Large HDL2 is superior to small HDL3 in the inverse association with CAD. The combination of HDL2, MCP-1, and hsCRP with CRFs provides an optimal prediction for CAD. 2021 Annals of Translational Medicine. All rights reserved.

Introduction

High-density lipoprotein-cholesterol (HDL-C) levels have been demonstrated to be inversely associated with coronary artery disease (CAD) risk in numerous epidemiological studies, including the Framingham Heart study (1-3). However, several recent clinical trials have challenged the value of raising HDL-C pharmacologically (4,5), and Mendelian randomization studies also do not support the causal role of HDL-C in CAD development (6). Instead of HDL “quantity”, growing interest has focused on the value of HDL “quality”, such as HDL subfractions and their differential atheroprotective functions (7).

HDL can be separated into two principal subfractions with ultracentrifugation: large buoyant HDL2 and small dense HDL3. Previous studies have suggested that HDL subfractions might add information in CAD risk assessment and Hirano et al. (8) developed a simpler and more precise method using a single-step precipitation to measure the levels of HDL2 and HDL3. HDL2 is often thought to be the “protective” form of HDL, and has been shown to have an inverse relationship with the risk of CAD in both cross-sectional and prospective studies (9,10). Our previous study has also consistently confirmed the inverse relationship of large HDL2 with premature CAD risk (11). However, conflicting results suggest that HDL3 is superior to HDL2 in predicting the incidence and mortality of ischaemic heart disease (12,13), and it remains to be determined which of the two is the more powerful negative risk factor for CAD and whether HDL subfractions provide a better predictive value than HDL-C itself.

Chronic persistent inflammation is another critical mechanism in the atherothrombotic process, and a substantial proportion of unexplained vascular disease might be related to inflammation (14). Monocyte chemoattractant factor-1 (MCP-1) is a chemoattractant that recruits monocytes to the subendothelial space and initiates plaque formation. Georgakis et al. (15) have shown that genetically determined higher MCP-1 levels are associated with a higher risk of CAD, myocardial infarction, and stroke, while the prospective PRIME study did not confirm these same associations (16). High-sensitivity C-reactive protein (hsCRP) is another inflammatory marker that displays a positive and independent relationship with cardiovascular morbidity and mortality (17,18). However, adding hsCRP levels to existing CAD risk assessment models demonstrates only modest improvement (19). Thus, the roles of MCP-1 and hsCRP in CAD risk prediction still require further research.

Since 1998, the original Framingham risk scores (FRS) (20) have incorporated age, sex, diabetes, systolic and diastolic blood pressures, levels of total and HDL cholesterol, and smoking to estimate a model for predicting the 10-year coronary heart disease risk and guide therapy for primary prevention. Considering that the original FRS ignored the risk prediction of other common cardiovascular disease (CVD), such as stroke, an FRS was developed in 2008 for the prediction of CVD (21). However, these classic tools have been shown to overestimate cardiovascular events by 8% to 154% in the MESA trial (22), calling for a combination of new risk factors to improve its performance. Ridker et al. (23) revealed that the combined evaluation of both LDL-C and hsCRP was superior as a method of cardiovascular risk detection than the measurement of either biologic marker alone, but little is known about the combined value of HDL subfractions with inflammatory markers in CAD risk assessment.

Therefore, we conducted this cross-sectional study, which enrolled 245 non-CAD subjects with low/moderate/high Framingham risk (L/M/H-FR) and 180 CAD patients, and used the new single-step precipitation method (8) to measure HDL2 and HDL3 levels. We found both HDL2 and HDL3 gradually decreased from the L/M/H-FR group to the CAD group. Large HDL2 is superior to small HDL3 in the inverse association with CAD. Furthermore, the combination of HDL2, MCP-1, and hsCRP with CRFs provides an optimal prediction for CAD risk. We present the following article in accordance with the MDAR checklist (available at http://dx.doi.org/10.21037/atm-21-948).

Methods

Study design and population

The study complied with the Declaration of Helsinki (as revised in 2013) and was approved by the hospital’s ethics review board (Sun Yat-sen Memorial Hospital, Guangzhou, China, IRB number SYSEC-KY-KS-2020-083-001). All subjects provided written informed consent to participate in the study. The study had been registered in Chinese Clinical Trial Registry as “A study on the predictive value of inflammatory markers and blood lipid profiles in the risk of coronary artery disease”. The clinical trial registration number is ChiCTR2000038859.

From September to November 2020, 245 asymptomatic subjects with no known CAD were recruited from our physical examination centre. According to the Framingham risk score [1998] (20), these participants were stratified into three groups: low-Framingham risk (L-FR, 10-year CAD risk <10%, n=120), moderate-Framingham risk (M-FR, 10%≤ 10-year CAD risk <20%, n=51) and high Framingham risk (H-FR, 10-year CAD risk ≥20%, n=74), and all of these participants were defined as the Framingham risk group (FR group). Meanwhile, 180 patients with angiographically diagnosed CAD (at least 50% obstructive lesions of one or more coronary arteries) in our cardiovascular medicine department were enrolled as the CAD group.

Subjects were excluded if they were suffering from acute coronary syndrome, advanced congestive heart failure, arrhythmia, or any infectious or systematic inflammatory disease over the previous 1 month. Other exclusion criteria included subjects less than 18 years or over 80 years, pregnancy or lactation, the existence of structural heart disease or cardiomyopathy, severe liver or renal dysfunction, thyroid disease, significant haematologic disorders, malignancy, major surgery or severe trauma within 2 months, and a history of the use of lipid-lowering drugs, antioxidants, glucocorticoid, or nonsteroidal anti-inflammatory drugs within the past 3 months. All exclusion criteria were preestablished.

Hypertension was defined as repeated BP measurements ≥140/90 mmHg (at least two times on different days) or self-reported hypertension with current use of anti-hypertension drugs. Diabetes mellitus was defined as fasting plasma glucose ≥126 mg/dL (7.0 mmol/L), self-reported diabetes, or use of anti-diabetic medications. A family history of premature CAD (PCAD) was defined as myocardial infarction occurrence before the age of 55 for men and 65 for women in first-degree relatives.

Measurement of HDL-C subfractions and other lipid profiles

Fasting blood samples were collected in EDTA tubes at baseline. After centrifugation at 4 °C, all plasma aliquots were stored at −80 °C until analysis.

The levels of HDL-C were measured by homogeneous HDL-EX HDL-C assays (Denka Seiken, Tokyo, Japan) and HDL subfractions were analysed by a single-step precipitation method (8). The precipitation reagent (0.06 mL), containing heparin, MnCl2, and dextran sulfate, was added to 0.3 mL of serum, mixed, left at room temperature for 30 min, and centrifuged at 10,000 rpm for 10 min at 4 °C. Both the apolipoprotein B (apo B)-containing lipoproteins and HDL2 could be simultaneously precipitated, and an aliquot of the supernatant was taken for HDL3 measurement using homogenous HDL-EX HDL-C assays. There were biological replicates for three times when testing the levels of HDL-C and HDL3. The measured value for total HDL3 was multiplied by 1.2 to correct for dilution by the reagents and levels of HDL2 were derived from the formula: HDL2 = HDL-C − HDL3.

Other lipid profiles, including triglyceride (TG), total cholesterol (TC), low-density lipoprotein-cholesterol (LDL-C), and apolipoprotein A1 (apo A1) and apo B, were measured on the Hitachi Modular Analytics System (Roche Modular DPP; Hitachi Ltd, Tokyo, Japan). Non-HDL-C levels were calculated by subtracting HDL-C from TC.

Measurement of inflammatory markers

The levels of plasma MCP-1 were measured by ELISA (BMS281TEN, Austria) on an automated microplate reader (Multiskan MK3, Thermo Fisher Scientific, USA). High-sensitivity testing for C-reactive protein (CRP) was performed using an immunoturbidimetric assay (GS621M, China) on the Hitachi Modular Analytics System.

Statistical analysis

Continuous variables with a Gaussian distribution were reported as the means ± standard deviations, and their differences between the FR group and CAD group were assessed by independent t-test. Continuous variables without a Gaussian distribution were reported as medians with interquartile ranges with differences compared by the Mann-Whitney U test. Categorical variables were reported as case numbers (proportions), and differences were compared using the Chi-squared test. Accordingly, the differences among the L/M/H-FR and CAD groups were compared by one-way ANOVA, the Kruskal-Wallis H test, or the Chi-squared test. The pairwise comparisons among the four groups were conducted by the Tukey or Game-Howell test for Gaussian-distributed variables and by the Kruskal-Wallis H test for others without a Gaussian distribution.

Multivariable logistic regression analysis was performed to estimate the CAD risk models. All models were adjusted for conventional risk factors (CRFs), including age, sex, body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), hypertension, family history of PCAD, smoking, fasting blood glucose (FBG), creatinine (CREA), TG, TC or LDL-C, apo A1 and apo B, which were based on univariate analysis and general knowledge of CAD risk. Models 1 to 8 referred to the model applied with a single factor of HDL parameters (HDL-C, HDL2, HDL3, HDL2/HDL3, HDL2/HDL-C, and HDL3/HDL-C) or inflammatory markers (MCP-1 and hsCRP) sequentially. Model 9 was applied with all six HDL parameters, and model 10 incorporated all six HDL parameters and the two inflammatory markers together. Odds ratios (ORs) with 95% CIs were calculated for these variables and shown in forest plots.

The discrimination of every model was assessed by the C-statistic, which was analogous to the area under the receiver operating characteristic (ROC) curve. The Z value, the index for the difference between two C-statistics, was applied to compare discrimination between the two models. Reclassification improvement was defined as an increase in risk category for individuals who suffered events and as a decrease for those who did not (21), and both net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used for comparison of reclassification between two models. When calculating NRI, clinically based cut-off points of 10% and 20% were used (21).

All statistical analyses were performed by SPSS software (version 25.0, SPSS, Chicago, Illinois, USA) and R version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria). A two-tailed P<0.05 was considered statistically significant.

Results

Baseline characteristics

The baseline demographic and clinical characteristics of the FR and CAD groups are shown in .

Table 1

Baseline characteristics in the Framingham risk group and CAD group

Characteristics	Framingham risk group (n=245)				CAD group (n=180)	P value#	P value§
	Total (n=245)	Low-risk (n=120)	Moderate-risk (n=51)	High-risk (n=74)
Age, years	59.77±10.90	52.23±9.51	66.37±6.35	67.46±6.16	60.09±9.57	0.746a	<0.001α*
Male (%)	136 (55.5)	60 (50.0)	30 (58.8)	46 (62.2)	98 (54.4)	0.827c	0.381ε
BMI, kg/m2	23.29±1.90	22.83±1.28	23.74±2.43	23.75±2.16	23.69±2.62	0.088a	<0.001α*
SBP, mmHg	120.0 (118.0–130.0)	118.0 (115.0–120.0)	120.0 (115.0–132.0)	130.0 (125.0–150.0)	140.0 (130.0–168.8)	<0.001b*	<0.001β*
DBP, mmHg	75.0 (70.0–80.0)	73.5 (70.0–75.0)	75.0 (70.0–80.0)	80.0 (70.0–90.0)	80.0 (75.0–100.0)	<0.001b*	<0.001β*
Hypertension (%)	67 (27.3)	9 (7.5)	19 (37.3)	39 (52.7)	64 (35.6)	0.070c	<0.001ε*
Diabetes (%)	0 (0)	0 (0)	0 (0)	0 (0)	64 (35.56)	<0.001c*	<0.001ε*
FPCAD (%)	2 (0.8)	2 (1.7)	0 (0)	0 (0)	10 (5.6)	0.004c*	0.044ε
Smoking (%)	47 (19.2)	9 (7.5)	5 (9.8)	33 (44.6)	36 (20.0)	0.834c	<0.001ε*
Drinking (%)	7 (2.9)	2 (1.7)	3 (5.9)	2 (2.7)	8 (4.4)	0.381c	0.411ε
FBG, mmol/L	4.70 (4.30–5.30)	4.55 (4.30–5.08)	4.90 (4.50–5.40)	5.05 (4.50–5.50)	5.41 (4.91–6.20)	<0.001b*	<0.001β*
CREA, μmol/L	96.34±16.69	86.80±13.43	105.10±15.29	105.77±13.55	97.58±12.91	0.389a	<0.001α*
UA, μmol/L	312.02±56.90	297.53±46.73	316.24±62.01	332.59±61.96	315.65±65.11	0.549a	<0.001α*

P value# represents the comparison of the Framingham risk (FR) group and CAD group; a, represents Student’s t-test; b, represents the Mann-Whitney U test; c, represents the Chi-squared test. P value§ represents the comparison of the low/moderate/high-FR group and CAD group; α, represents one-way ANOVA; β, represents Kruskal-Wallis H test; ε, represents Chi-squared test; *, P<0.05 as significance. CAD, coronary artery disease; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; FPCAD, family history of premature coronary artery disease; FBG, fasting blood glucose; CREA, creatinine; UA, uric acid.

Participants in both FR and CAD groups were mainly middle-aged to elderly, and slightly over half were men. In comparison with the FR group, patients in the CAD group had higher SBP and DBP, a higher percentage of diabetes, and positive family history of PCAD. FBG was higher in the CAD group, whereas CREA and uric acid (UA) did not show significant differences between the two groups. Furthermore, compared with the Low- and Moderate-FR groups, the High-FR group was more likely to have older subjects, more males, subjects with higher SBP levels, and a higher percentage of hypertension, diabetes, and smoking status.

Analysis of HDL-C subfractions

shows HDL2 accounted for approximately two-fifths of HDL-C, and HDL3 accounted for the other three-fifths. All HDL-C, HDL2, and HDL3 levels were lower in the CAD group than in the FR group (HDL-C: 1.07±0.14 vs. 1.27±0.20; HDL2: 0.40±0.07 vs. 0.51±0.09; HDL3: 0.67±0.12 vs. 0.76±0.11; all P<0.001, respectively).

The distributions of both HDL2 and HDL3 were consistent with HDL-C, with a reduced level across the L/M/H-FR and CAD groups (HDL-C: 1.38±0.15 vs. 1.24±0.18 vs. 1.14±0.18 vs. 1.07±0.14; HDL2: 0.56±0.06 vs. 0.50±0.07 vs. 0.44±0.10 vs. 0.40±0.07; HDL3: 0.81±0.09 vs. 0.74±0.11 vs. 0.70±0.11 vs. 0.67±0.12; all P<0.001, respectively). However, only HDL-C and HDL2, but not HDL3, HDL2/HDL3, HDL2/HDL-C, or HDL3/HDL-C showed significant differences in pairwise comparisons among all four groups ().

Figure 1

Pairwise comparisons of HDL parameters and inflammatory markers between the Low/Moderate/High Framingham risk group and the CAD group. (A) HDL-C and its subfraction concentrations; (B) ratios of HDL parameters; (C) MCP-1 concentration; (D) hsCRP concentration. *P<0.05; ***P<0.001. P<0.05 was considered significant. HDL, high-density lipoprotein; CAD, coronary artery disease; MCP-1, monocyte chemoattractant protein-1; hsCRP, high-sensitivity C-reactive protein.

Analysis of inflammatory markers

As shown in , both MCP-1 and hsCRP levels were significantly higher in the CAD group than in the FR group [MCP-1: 56.09±13.16 vs. 39.60±12.45; hsCRP: 4.24 (2.56–10.48) vs. 1.53 (0.77–2.84); P<0.001, respectively] and also showed a gradual increase from the L/M/H-FR group to the CAD group. The difference of MCP-1 was not significant between the L-FR and M-FR groups but was significant in pairwise comparisons among the M-FR, H-FR, and CAD groups (all P<0.05, ). The level of hsCRP was significantly different in the comparisons between the L-FR and M-FR groups as well as between the H-FR and CAD groups (P<0.001, respectively) but was similar between the M-FR and H-FR groups (P>0.05) ().

Table 2

Biochemistry parameters of lipid profiles and inflammation in the Framingham risk group and CAD group

Variable	Framingham risk group (n=245)				CAD group (n=180)	P value#	P value§
	Total (n=245)	Low-risk (n=120)	Moderate-risk (n=51)	High-risk (n=74)
HDL parameters
HDL-C, mmol/L	1.27±0.20	1.38±0.15	1.24±0.18	1.14±0.18	1.07±0.14	<0.001a*	<0.001α*
HDL2, mmol/L	0.51±0.09	0.56±0.06	0.50±0.07	0.44±0.10	0.40±0.07	<0.001a*	<0.001α*
HDL3, mmol/L	0.76±0.11	0.81±0.09	0.74±0.11	0.70±0.11	0.67±0.12	<0.001a*	<0.001α*
HDL2/HDL3	0.67±0.09	0.69±0.04	0.68±0.06	0.63±0.13	0.61±0.16	<0.001a*	<0.001α*
HDL2/HDL-C	0.40±0.03	0.41±0.01	0.40±0.02	0.38±0.05	0.37±0.06	<0.001a*	<0.001α*
HDL3/HDL-C	0.60±0.03	0.59±0.01	0.60±0.02	0.62±0.05	0.63±0.06	<0.001a*	<0.001α*
Standard lipid profiles
TG, mmol/L	1.45±0.41	1.29±0.25	1.56±0.44	1.62±0.49	2.10±0.88	<0.001a*	<0.001α*
TC, mmol/L	4.72±0.98	4.05±0.53	4.80±0.53	5.75±0.89	5.46±1.03	<0.001a*	<0.001α*
LDL-C, mmol/L	2.93±0.97	2.25±0.49	3.01±0.59	3.96±0.83	3.87±0.96	<0.001a*	<0.001α*
Non-HDL-C, mmol/L	3.45±1.08	2.68±0.55	3.56±0.56	4.61±0.92	4.29±1.03	<0.001a*	<0.001α*
apo A1, g/L	1.30±0.17	1.39±0.15	1.24±0.12	1.19±0.14	1.14±0.14	<0.001a*	<0.001α*
apo B, g/L	0.91±0.12	0.88±0.12	0.93±0.10	0.95±0.12	1.04±0.69	0.013a*	<0.001α*
Inflammatory markers
MCP-1, pg/mL	39.60±12.45	36.51±11.28	38.64±9.18	45.26±14.29	56.09±13.16	<0.001a*	<0.001α*
hsCRP, mg/L	1.53 (0.77–2.84)	1.07 (0.65–1.65)	2.51 (0.96–3.86)	2.58 (1.28–3.89)	4.24 (2.56–10.48)	<0.001b*	<0.001β*

P value# represents the comparison of the Framingham risk (FR) group and CAD group; a, represents Student’s t-test; b, represents the Mann-Whitney U test. P value§ represents the comparison of the low/moderate/high-FR group and CAD group; α, represents one-way ANOVA; β, represents the Kruskal-Wallis H test; *, P<0.05 as significance. HDL-C, high-density lipoprotein-cholesterol; TG, triglyceride; TC, total cholesterol; LDL-C, low-density lipoprotein-cholesterol; apo A1, apolipoprotein A1; apo B, apolipoprotein B; MCP-1, monocyte chemoattractant protein-1; hsCRP, high-sensitivity C-reactive protein.

Multivariable logistic regression analysis

In models 1 to 8, which included a single parameter sequentially after adjustment for CRFs (), both HDL-C (OR 0.011, 95% CI, 0.002–0.071; P<0.001) and HDL2 (OR 0.000072, 95% CI, 0.000001–0.004; P<0.001) were independently associated with CAD risk in models 1 and 2, while HDL3 and three ratios of HDL parameters failed to be significantly related to CAD. In addition to inflammatory markers, MCP-1 (OR 1.082, 95% CI, 1.056–1.108; P<0.001) and hsCRP (OR 1.160, 95% CI, 1.068–1.260; P<0.001) demonstrated positive relationships with CAD risk in models 7 and 8.

Figure 2

CAD risk prediction models by HDL subfractions and/or inflammatory markers. Multivariable logistic regression analysis was performed to evaluate the associations of HDL subfractions and/or inflammatory markers with CAD risk. Odds ratios (ORs) were estimated with 95% confidence intervals (CIs); P valuea represent the significance of ORs. The C-statistics of every model were computed; the P valueb represents the significance of the C-statistics with P<0.05 in bold type as significance. a, refers to a model including a single HDL (HDL-C, HDL2, HDL3, HDL2/HDL3, HDL2/HDL-C or HDL3/HDL-C) or inflammatory (MCP-1 or hsCRP) parameter after adjusting for conventional risk factors (CRFs). b, refers to a model including all six HDL parameters together after adjusting for CRFs. c, refers to a model including all six HDL parameters and two inflammatory markers after adjusting for CRFs. HDL, high-density lipoprotein; CAD, coronary artery disease; MCP-1, monocyte chemoattractant protein-1; hsCRP, high-sensitivity C-reactive protein; CRF, conventional risk factor.

Further, when applying all six HDL parameters in model 9, only large HDL2 (OR 0.000072, 95% CI, 0.000001–0.004; P<0.001), rather than HDL-C or small HDL3, maintained the inverse relationship with CAD ().

Finally, when all six HDL parameters and two inflammatory markers were combined into model 10, HDL2 (OR 0.001, 95% CI, 0.000027–0.051; P<0.001), MCP-1 (OR 1.066, 95% CI, 1.039–1.094; P<0.001) and hsCRP (OR 1.130, 95% CI, 1.041–1.227; P=0.003) showed significance ().

Model fit and risk reclassification

Models 1, 2, and 7 to 10 were defined as significant models with significant ORs of HDL parameters and/or inflammatory markers (P<0.05), and their fits were compared by Z value, NRI, and IDI.

As shown in , model 2 with HDL2 had a higher C-statistic than model 1 with HDL-C (0.916 vs. 0.914). However, the discrimination of model 2 was virtually identical to that of model 1 with a nonsignificant Z value (0.595, P=0.552, ), and no reclassification improvement was observed with a negative NRI (0.011, P=0.731) and IDI (0.014, P=0.078) ().

Figure 3

Comparisons of significant multivariable logistic regression models for CAD risk assessment. Model 1, model 2, and models 7 to 10 were defined as significant models with significant ORs of HDL parameters and inflammatory markers (P<0.05). (A) Comparisons of Z value, NRI, and IDI of the six significant models. The Z value represents the comparison of C-statistics; P valuea represents the significance of the Z value. NRI, net reclassification improvement, was calculated with clinically based cut-off points for CAD risk as 10% and 20%; P valueb for significance of NRI. IDI, integrated discrimination improvement; P valuec for significance of IDI. (B) ROC curve for comparison of model 2 with model 1 in the C-statistic. (C,D,E,F) ROC curves for comparisons of model 10 with models 1, 2, 7, and 8 in the C-statistics. (G) ROC curve for comparison of model 10 with model 9 in the C-statistic. P<0.05 as significance. HDL, high-density lipoprotein; CAD, coronary artery disease; OR, odds ratio; NRI, net reclassification improvement; IDI, integrated discrimination improvement; ROC, receiver operating characteristic.

Model 10 was applied with significant HDL2, MCP-1, and hsCRP. Compared with the four significant models (models 1, 2, 7, and 8) that included a single factor sequentially, model 10 showed a significantly higher C-statistic (0.942 vs. 0.914, 0.916, 0.927 and 0.921, all P<0.05), respectively () and an improvement in reclassification with positive NRI and IDI (NRI: 0.203, 0.179, 0.105, 0.172, respectively; IDI: 0.095, 0.081, 0.049, 0.052, respectively; all P<0.05) (). Furthermore, in the comparison with model 9 incorporating all six HDL parameters, model 10 also showed improved fit on all measures with a Z value of 3.299 (P<0.001), NRI of 0.179 (P<0.001), and IDI of 0.081 (P<0.001) ().

Discussion

In the present study, we found that the levels of HDL-C, HDL2, and HDL3 gradually decreased while MCP-1 and hsCRP gradually increased from the L/M/H-FR group to the CAD group. Both HDL-C and large HDL2, but not small HDL3 or the ratios of HDL parameters, showed an inverse and independent association with CAD risk. Only HDL2 remained significant in the model including all six HDL parameters. Moreover, the combined model with HDL2, MCP-1, and hsCRP had better discrimination and risk reclassification than the models including either a single factor or all HDL parameters.

HDLs comprise a family of heterogeneous particles that vary by size, density, composition, and functionality. Two distinct HDL subfractions, large buoyant HDL2 and small dense HDL3, might exert differential effects on atherosclerosis and display a promising role in CAD risk prediction. In our study, large HDL2 levels, but not small HDL3 or three ratios of HDL parameters, showed significant differences in pairwise comparisons among the L/M/H-FR and CAD groups and remained inversely associated with CAD independent of CRFs. Consistent with our findings, previous data have yielded an inverse relationship between HDL2 levels and CAD (9,10,24), carotid atherosclerosis (7,25), and type 2 diabetes (26). Cholesterol efflux capacity (CEC) is crucial to HDL in the anti-atherogenic process. El Khoudary et al. (27) found a strong correlation of large-HDL subspecies with macrophage CEC in healthy women before and after menopause. In patients with coronary endothelial dysfunction, an early marker of atherosclerosis, the selective decrease in large-HDL concentrations had already contributed to impaired cholesterol efflux from endothelial cells (28). However, a more protective effect of HDL3 over HDL2 (29) or equal benefits of both subfractions for CAD (30) has also been reported. Martin et al. (13) have also revealed that the central positioning and inefficiency of HDL3 in cholesterol efflux potentially explained the trend of lower HDL3 with a higher risk for mortality/myocardial infarction in people with secondary prevention. This inconsistency may be caused by the differences in the health status of the populations studied, the methods used to subclassify HDL-C, and perhaps the anti-atherosclerotic property of various HDL particles.

In addition, although the model with HDL2 yielded no fit improvement over the model with HDL-C, only HDL2 showed significance when applying all six HDL parameters in the CAD risk model with adjustment for CRFs. This suggests a possible superiority of HDL2 to HDL-C in CAD risk prediction. HDL2 has been proven to be more closely related to carotid intima-media thickness than total HDL-C in healthy middle-aged individuals (31). Lamon-Fava et al. (32) also found that the power of large-HDL particles quantified by 2D gel electrophoresis to predict coronary atherosclerosis risk was typically superior to that of HDL-C in postmenopausal women. However, some studies elucidated that neither HDL2 nor HDL3 was better at CAD risk prediction than HDL-C itself (33). The superiority of HDL2 over HDL-C requires verification in a large-scale prospective study.

Inflammation is ubiquitous in the atherothrombotic process, and inflammatory biomarkers, including MCP-1 and hsCRP, were positively related to CAD risk independent of CRFs in our study. MCP-1 is likely to play a pathogenic role in CAD by recruiting monocytes to initiate plaque formation and activating the ubiquitin-proteasome system to trigger plaque rupture (34). Therefore, its levels may reflect increased atherosclerotic burden, enhanced plaque vulnerability, or both (35). We also found that MCP-1 levels were similar in the L- and M-FR groups but showed a significant difference between the M- and H-FR groups, indicating that MCP-1 could be applied in discriminating persons with high CAD risk among asymptomatic subjects for more aggressive intervention.

Different from MCP-1, hsCRP is predominantly a biomarker of inflammation rather than a causal factor (36). Increased hsCRP levels were associated with an increased risk of CAD and ischaemic stroke (17) and adds as much to risk prediction as total cholesterol, HDL-C and blood pressure (18). The difference in hsCRP levels was significant between the L- and M-FR groups but lost significance when comparing M-FR with the H-FR group, which could be explained by hsCRP being an acute reactant that responds easily and quickly to low-grade inflammation while maintaining a stable level in persistent inflammation (35).

The incidence of CAD is strongly associated with a set of traditional risk factors, and the inclusion of new risk factors may improve the performance of current multivariable risk assessment tools. By incorporating hsCRP and a family history of premature CAD, the Reynolds Risk Score shows improved calibration and discrimination over the Framingham cardiovascular risk score and the Adult Treatment Panel III score (37). Therefore, we applied all six HDL parameters and two inflammatory markers into the CRFs-adjusted model, in which HDL2, combined MCP-1 and hsCRP, showed significance. Impressively, this model has achieved significant improvement in discrimination and risk reclassification over the models including either a single factor alone or all six HDL parameters together. Canoui-Poitrine et al. (16) also added two inflammatory biomarkers, RANTES and IP-10, to the traditional risk factor-based model for ischaemic stroke risk and gained improvement in both C-statistics and reclassification. However, the combined value of HDL subfractions and inflammatory markers needs to be evaluated in further prospective studies. To an extent, our data suggest that the combined screening of HDL subfractions with inflammatory biomarkers may provide better predictive information than screening for either a biomarker alone or all six HDL parameters together.

Limitations

First, this was a cross-sectional study, and the results should be confirmed in prospective trials. Additionally, the study was conducted in a single centre with a relatively small sample size, and conflict existed regarding whether HDL2 was superior to HDL-C in CAD risk assessment. Further exploration is needed in a larger-scale study.

Conclusions

In conclusion, the large HDL2 is superior to small HDL3 in the assessment of CAD risk. In the model including all six HDL parameters, only HDL2, but not HDL-C, maintained the inverse and independent association with CAD, revealing the possible superiority of HDL2 to HDL-C. The combination of HDL2, MCP-1, and hsCRP with CRFs provides an optimal and validated prediction for CAD than a single biomarker alone or all HDL parameters together.

The article’s supplementary files as