Lipoprotein subfractions and coronary artery calcium progression: insights from the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil)

Atherosclerosis is a chronic inflammatory disease that often remains clinically silent until advanced stages, and traditional lipid biomarkers provide limited information on atherosclerotic burden and residual cardiovascular risk.1 Longitudinal evidence linking lipoprotein subfractions, particularly high-density lipoprotein (HDL)-related particles, to coronary artery calcium (CAC) progression remains sparse.2 CAC scoring is an established standard for quantifying subclinical atherosclerosis and predicting cardiovascular events,3 yet particle-based profiling may offer incremental value.4 We evaluated whether lipoprotein subfractions quantified by nuclear magnetic resonance (NMR) spectroscopy associate with CAC progression in the ELSA-Brasil cohort. This ancillary longitudinal study analysed 689 participants from ELSA-Brasil5 with CAC assessed at baseline (2008–2010) and follow-up (2012–2013; mean interscan interval: 4.8 years). Participants were civil servants aged 35–74 years from six Brazilian cities with baseline CAC >0. CAC progression was defined by the validated Berry method,6 based on the square root difference between follow-up and baseline Agatston scores, and quantified by semi-automated software (Calcium Scoring; Philips Workstation) by a blinded reader. Participants with extreme annual CAC changes (>400 Agatston units/year or >99th percentile) were excluded from the main analysis and re-included in sensitivity analysis (n = 720). Twenty-five lipoprotein subfractions were quantified by proton NMR spectroscopy (LipoProfile® 4, LabCorp; intra-assay coefficient of variation <5% for particle concentrations, <2% for size), encompassing triglyceride-rich particles, cholesterol-rich low-density lipoprotein (LDL) particles, cholesterol-rich HDL particles and subspecies (in nanometres), and glycoprotein acetylation (GlycA, in µmol/L). Exposures were analysed categorically (highest quartile vs. remainder) and continuously (per 1-standard deviation [SD] increment), with P-values false discovery rate (FDR)-adjusted (Benjamini–Hochberg) across the 25 exposures. Missing covariates were handled by multiple imputation by chained equations (five datasets). Two stabilized inverse probability weighting (IPW) models were constructed: Model 1 excluded triglycerides, HOMA-IR, and lipid-lowering therapy as potential intermediates; Model 2 included these covariates. Weights were estimated by logistic regression, stabilized by the marginal probability of progression with no truncation; positivity was assessed by weight overlap. Covariate balance was confirmed (all standardized mean differences <0.10; effective sample sizes: 620.4 and 610.9; stabilized weight ranges: 0.40–3.19 and 0.40–3.37) (Figure 1, Panels A, B). Metabolic syndrome was defined by harmonized criteria (≥3 of the following: abdominal obesity, hypertension, hyperglycaemia, low HDL cholesterol, hypertriglyceridaemia; medication use counted as meeting each criterion). The study was approved by all Institutional Review Boards (CAAE: 0016.1.198.000-06); all participants provided written informed consent.