Abstract
Objectives
Cardiovascular disease (CVD) is the leading cause of premature morbidity and mortality among African American women, and diet plays a crucial role in its prevention. Diet consists of a complex mixture of foods and nutrients, yet few existing statistical methods can account for potential nonlinear and interactive relationships between multiple dietary factors and their effects on health. To realistically assess dietary impacts on CVD risk among African American women, we utilized an innovative statistical approach, Bayesian Kernel Machine Regression (BKMR), which takes into consideration the relationship between multiple dietary factors, as mixtures and as individual components, and CVD risk.
Methods
Using data from 2724 healthy African American participants of the Women's Health Initiative Observational Study, we examined the association of nine dietary factors (fruits, vegetables, fish, red meat, poultry, nuts, whole grains, dairy, and sodium), collected through a validated food frequency questionnaire, with both systolic blood pressure (SBP) and CVD incidence. Through a kernel machine representation, BKMR regresses the outcome on a smooth function of the exposures, adjusting for potential confounders and allowing for possible nonlinearities and interactions. We used BKMR for modeling the continuous outcome, SBP, and its probit extension for the binary outcome, CVD incidence.
Results
Whole grain and fish had the strongest associations with SBP. SBP decreased by 0.78 mmHg (95% credible interval (CI): −1.70, 0.14) and increased by 0.70 mmHg (95% CI: −0.12, 1.52) for an interquartile range (IQR) increase in whole grain and fish consumption, respectively. We saw a linear and increasing association between the diet mixture and CVD incidence. This trend was mainly driven by red meat consumption as the primary dietary risk factor to CVD incidence: an IQR increase in red meat consumption was associated with a 0.06-unit (95% CI: −0.02, 0.14) increase in the probit CVD risk. No evidence for interactions and nonlinearities was observed.
Conclusions
BKMR is a novel method for modeling complex dietary mixtures by incorporating potential nonlinearities and interactions, allowing identification of major dietary factors associated with elevated SBP and CVD incidence among a population disproportionally affected by CVD.
Funding Sources
NHLBI.