Cox Regression analysis was used to calculate the multivariable adjusted hazard ratios in different levels of chocolate intake. In order to account for changes over time in the predictor, chocolate consumption, the chocolate consumption values at exam 1 were updated at exam 3 with available chocolate consumption values. A time-dependent covariates analysis8 was used to achieve the updating. A sensitivity analyses was conducted using only the actual chocolate intake at exam 1 as predictor e without updating at exam 3. Exclusion criteria were used to help prevent a bias caused by the
tendency of ARIC participants to decrease their chocolate intake after being diagnosed with a serious chronic disease that includes diabetes, heart attack, stroke and cancer.9 All exam 1 data were excluded for participants who at the exam reported a prior diagnosis of heart attack, stroke or cancer, or who had a diagnosis of diabetes based on verified usage of anti-diabetic medications.