Leukocyte DNA methylation-based signatures for atherosclerotic cardiovascular disease risk prediction in the Million Veteran Program
TL;DR
DNA methylation-based signatures were associated with incident ASCVD and modestly improved risk prediction beyond that of traditional risk factors, and an agnostic probe reliability-based approach was developed.
Abstract
Background and Aims: The contribution of DNA methylation signatures to atherosclerotic cardiovascular disease (ASCVD) risk prediction remains unclear. We developed methylation risk scores (MRS) for incident ASCVD and assessed whether they improved risk prediction beyond established risk factors. Methods: We studied 44,674 Million Veteran Program participants with leukocyte DNA methylation data, divided into two independent subcohorts: a prevalent ASCVD cohort (n=27,560) used for epigenome-wide association analyses (EWAS) to inform cytosine-phosphate-guanine dinucleotide selection, and a cohort free of ASCVD at blood draw (n=17,114), split into training and testing sets for MRS development and evaluation. MRS for incident ASCVD were developed using elastic net regression. Incremental prediction beyond clinical risk factors was assessed by improvement in discrimination ({Delta}CPE), reclassification (NRI), and calibration. Results: Three MRS were developed: MRS-1A, informed by prevalent ASCVD EWAS and probe reliability; MRS-1B, informed by EWAS alone; and MRS-2, using an agnostic probe reliability-based approach. Among 17,114 participants (mean [SD] age, 58.9 [14.1] years; 89.6% men; 54.2% European), 2,789 developed ASCVD over a median follow-up of 7.4 years. Each MRS was associated with incident ASCVD (HR per 1-SD: 1.97 [95% CI, 1.72-2.26] for MRS-1A, 2.08 [1.83-2.37] for MRS-1B, and 2.07 [1.78-2.39] for MRS-2) and modestly improved discrimination beyond clinical risk factors ({Delta}CPE: 0.014 [0.006, 0.021], 0.016 [0.007, 0.023], and 0.013 [0.006, 0.021], respectively). MRS improved risk stratification, driven by the downward reclassification of non-events (non-event NRI: 3.6% [2.6-4.7], 5.4% [4.3-6.5], and 3.6% [2.6-4.6], respectively), while maintaining calibration. Conclusions: DNA methylation-based signatures were associated with incident ASCVD and modestly improved risk prediction beyond that of traditional risk factors.