Study participants were multi-ethnic men and women ages 45-84 from MESA, without baseline clinical CVD or lipid-lowering medication use (). HDL-C and HDL-P concentrations were positively correlated (). HDL-C and HDL-P were inversely correlated with LDL-C, weakly, (ρ= −0.08, −0.13, respectively), and LDL-P, more strongly, (ρ= −0.38, −0.25) and with other metabolic risk factors (e.g., small LDL-P, triglycerides, BMI, waist circumference, and HOMA-IR), but for all, correlations were stronger for HDL-C than for HDL-P ().
Characteristics of MESA Participants in Study (n=5598), 2000-2002
Spearman correlations of the concentrations of HDL cholesterol (HDL-C) and HDL particles (HDL-P) among MESA participants in study (n=5598), 2000-2002
LDL-C and LDL-P
Positive associations of LDL-C and LDL-P with cIMT and CVD events in MESA have been published(21
) and when LDL-P and LDL-C differ, associations are stronger for LDL-P(23
). In our study, adjusted for base covariates, the cIMT difference (95%CI) per 1-SD increment was 28.8(13.4, 44.3) μm for LDL-C, and 36.5 (20.8, 52.1) μm for LDL-P. The HR (95%CI) for CHD per 1-SD increment was 1.24 (1.09, 1.42) for LDL-C and 1.29 (1.13, 1.47) for LDL-P, adjusted for base covariates. LDL-C and LDL-P each remained associated (p<0.05) with cIMT and incident CHD in the models reported below that also adjusted for HDL-P, HDL-C or both.
Associations with Carotid Atherosclerosis
Adjusted for base covariates (age, sex, ethnicity, hypertension and smoking) mean cIMT was lower with higher quartiles of HDL-C (, Panel A) or HDL-P (, Panel B). The inverse linear association between HDL-C and cIMT was attenuated by adjusting for LDL-P or for HDL-P, and was abolished when adjusted for both. In contrast, HDL-P remained inversely associated with cIMT after adjusting for LDL-P, HDL-C, or both (p for trend <0.05 for all).
Mean cIMT (μm) across quartiles of HDL-C or HDL-P, before and after adjusting for LDL-P and each other (n=5541).
Additional models are shown in , which reports adjusted mean cIMT difference for a 1-SD increment in HDL-C or HDL-P. Separately, higher HDL-C and HDL-P were each associated with lower mean cIMT, and were modestly attenuated by adjusting for LDL-C or HDL particle size. However, adjusting for LDL-P substantially attenuated cIMT estimates, more for HDL-C (12.2 (−21.4,−3.1) μm), than for HDL-P (−20.7 (−29.6, −11.8) μm). Further adjustment for LDL-C and triglycerides had little effect.
Predicted difference in carotid intima-media thickness (μm) for a 1 standard deviation (SD) greater HDL cholesterol (HDL-C) or HDL particle (HDL-P) concentration among MESA participants (n=5541), 2000-2002
In joint models, (HDL-C and HDL-P adjusted for each other and base covariates; lower panel), HDL-C and HDL-P associations were only mildly attenuated by adjusting for LDL-C. However, adjusted for HDL-P, HDL-C was no longer significantly associated with cIMT (−11.1 (−22.7, 0.42) μm, and became positive, but not statistically significant, in models that also adjusted for mean HDL size or LDL-P, with or without LDL-C and triglycerides. Conversely, HDL-P remained inversely associated with cIMT (−22.2 (−33.8, −10.6) μm) when adjusted for HDL-C, and also for LDL-P, or HDL size, or LDL-P, LDL-C and triglycerides.
To visualize the multivariable model results, shows adjusted mean cIMT for cross-classified tertiles of HDL-C and HDL-P (), and further stratified by median LDL-P (). Power is limited for these analyses due to small numbers in the discordant groups, e.g., high HDL-C/low HDL-P. Within HDL-P tertiles, HDL-C was generally not inversely associated with cIMT (p HDL-C trend= not significant for all) (), but within HDL-C tertiles, HDL-P was inversely associated with cIMT in both low and high HDL-C tertiles (p trend <0.05 for both). When further stratified by median LDL-P (), trends were generally not statistically significant, but HDL-C was positively associated with cIMT within 5 of the 6 HDL-P tertiles, whereas HDL-P remained inversely associated with cIMT within 5 of the 6 HDL-C tertiles.
Mean cIMT (μm) by joint tertiles of HDL-C and HDL-P, entire group, and stratified by median LDL-P (1236.5 nmol/l). All models adjusted for base covariates.
Sensitivity Analyses for cIMT Associations
Interaction terms for sex, ethnicity, diabetes status, or hsCRP were not significant. In sex-stratified models, adjusted for LDL-P and each other, HDL-C was not inversely associated with cIMT for women: −1.2 (−15.0, 12.7) μm, or men: 8.3 (−13.4, 30.1) μm, whereas HDL-P was significantly inversely associated with cIMT for women: −17.8 (−31.4, −4.2) μm, and men: −27.1 (−47.7, −6.4) μm. Evaluated separately, the internal and common carotid artery showed similar results to our combined cIMT measure, although associations were slightly stronger for the internal carotid artery, which is more prone to plaque.
Associations with Incident Coronary Heart Disease Events
Among the 5597 participants with incident CHD data, 227 CHD events occurred during mean(SD) =6.0(1.4) years follow-up. The proportionality assumption appeared valid, i.e., interactions of time with HDL-C or HDL-P were not significant. CHD risk is reported for quartiles and for a 1-SD increment in HDL-C and HDL-P, for separate and joint models (). Separately, higher HDL-C and HDL-P were similarly associated with lower CHD risk, adjusted for base covariates. Adjusting for LDL-C or HDL particle size had little effect. Adjustment for LDL-P attenuated the association for HDL-C to a greater extent than for HDL-P. In joint models (adjusted for HDL-P), HDL-C HRs were not statistically significant and became weakly positive when adjusted for LDL-P, LDL-C, and (log) triglycerides. In contrast, HDL-P’s inverse association with CHD remained statistically significant when adjusted for HDL-C, LDL-P, LDL-C, and triglycerides.
Risk of incident CHD across higher HDL-C and HDL-P quartiles and for 1-SD increment among MESA participants (n=5597), baseline 2000-2002
Sensitivity Analysis for CHD Associations
In multivariable models, interaction terms for sex, ethnicity, diabetes or hsCRP were not statistically significant. Results were similar if adjusted for baseline diabetes status, or if current hormone users were excluded, or if stratified by sex. With few cases among women (n=66), CHD risk estimates for women were not statistically significant, but the base covariate-adjusted HR (95%CI) for (1-SD) higher HDL-C (0.81 (0.63, 1.05) became 1.00 (0.71, 1.43) when adjusted for HDL-P and LDL-P, whereas for HDL-P, the base covariate-adjusted HR(95%CI) of 0.77 (0.59, 1.00), became 0.81 (0.57, 1.14) when adjusted for HDL-C and LDL-P. Results were also similar for the secondary outcomes of all CVD or hard CHD or hard CVD (Online Table 1
Finally, we evaluated CHD risk associations at very high levels of HDL-C and HDL-P. Adjusted for base covariates, LDL-P, HDL-P, and log-triglycerides the HR(95%CI) for very high HDL-C (≥80 mg/dl) compared with low HDL-C (<40 mg/dl) became positive and statistically significant (2.59 (1.11, 6.02). Conversely, adjusted for base covariates, LDL-P, HDL-C and log-triglycerides, the HR(95%CI) for analogous very high vs. low HDL-P (≥45.7 vs. <29 μmol/l) was 0.50 (0.19, 1.35).