Abstract MP31: Blood DNA Methylation Signatures of Incident Coronary Heart Disease: An Epigenome-wide Analysis in the Strong Heart Study

Background: In the US, American Indians suffer a disproportionate burden of CHD compared to other racial/ethnic groups. Additional strategies are needed to identify individuals at risk. Objectives: Investigate the association of blood DNA methylation (DNAm) with incident CHD in the Strong Heart Study and the prediction ability of DNAm beyond traditional risk factors. We maximized prediction ability using Bayesian Hierarchical Cox (BHCox) and Survival Random Forest models, which allow large numbers of CpGs in a single model, instead of considering them individually. Methods: Among 2325 men and women 45-74 years old in 1989-1991, 557 CHD events were identified over 20 years of follow-up. DNAm was measured in 790,026 CpGs, pre-processed and corrected for batch effects. We ran adjusted BHCox models for subsets of CpGs selected from similarly adjusted Cox models for individual CpGs. Prediction ability of CpGs in BHCox was further evaluated with Survival Random Forest, which is robust to overfitting. We also conducted a targeted analysis using Cox regression. Results: 26 CpGs associated with CVD in previous studies were nominally associated with incident CHD in a targeted analysis. The cross-validated C index for the model with traditional risk factors was 0.703. In BHCox, further entering 30K CpGs in a single model, resulted in 231 CpGs being significantly associated with incident CHD, with a cross-validation C statistic of 0.855. In Survival Random Forest, further entering the 231 CpGs from the BHCox model resulted in a cross-validation C statistic of 0.771 and 182 CpGs with variable importance (VIMP) greater than zero. The top CpGs in BHCox were also VIMP>0 in random forest models and were located in SLC24A1 (calcium exchanger), SHBG (sex hormone binding globulin), and LINC00346 (non-protein coding RNA 346). Conclusions: We found novel CpG sites prospectively associated with incident CHD and confirmed signals from previous studies. DNAm might help to identify individuals at high risk of developing CHD.