Oral Session

Nutritional Epidemiology

OR06-02-19 - Joint Association of Multiple Dietary Components on Cardiovascular Disease Risk: A Machine Learning Approach

Presenting Author(s)

• YZ

  Yi Zhao, MSc

  Tufts University, Friedman School of Nutrition

Co-Author(s)

• GS

  Gitanjali Singh, PhD, MPH

  Friedman School of Nutrition Science & Policy, Tufts University
• EN

  Elena Naumova, PhD

  Tufts University, Friedman School of Nutrition

Objectives : Suboptimal diet is associated with substantial cardiovascular disease (CVD) burden globally. Recognizing that populations are exposed to a complex mix of dietary factors, the aim of this study was to examine the potentially complex non-linear non-additive relationships between multiple components of diet and 10-year CVD risk using a machine learning method.

Methods : We implemented Bayesian Kernel Machine Regression (BKMR) using an R package bkmr among a group of healthy, middle-aged participants from the Coronary Artery Risk Development in Young Adults (CARDIA) cohort. Exposures were defined as common components of diet that have probable or convincing evidence of association with CVD risk. Outcome was an individual’s predicted 10-year CVD risk calculated using the revised Pooled Cohort Equations for Estimating Atherosclerotic CVD Risk by the American College of Cardiology/American Heart Association. Ten-year CVD risk was modeled as a flexible kernel function of the exposure variables, adjusted for potential confounding factors.

Results : The mean age and predicted 10-year CVD risk of the 2193 participants were 45.75 (SD: 3.07) years old and 0.02 (SD: 0.02). We found a significant positive joint relationship between the multiple components of diet with CVD risk when all dietary factors were all above their 25th percentile. We also identified positive associations between whole grain, refined grain, red meat with CVD risk, and negative associations for vegetable and fruit. Furthermore, red meat, refined grain, whole grain and added sugar demonstrated potential non-linear relationships with the CVD risk. Our result also suggested potential interaction between refined grain with other dietary factors.

Conclusions : Machine learning is a promising tool to estimate the joint associations between multiple diet components and CVD risk. It is able to identify the joint, univariate dose-response relationship between diet and CVD risk as well as component-wise interactions.

Funding Sources :

This work was supported by a funding from National Heart, Lung, and Blood Institute (NHLBI).