Most physicians base decisions about cardiovascular treatment based on a calculation of overall cardiovascular disease risk.
Nutrition is not currently accounted for in cardiovascular risk calculators, despite being a major contributor to cardiovascular risk. Nutrition is commonly measured through 24-hour dietary recalls, which are difficult to summarize and condense into standard metrics to improve cardiovascular risk estimation.
We propose to use deep learning to study the large sparse matrix of nutrition data from the National Health and Nutrition Examination Survey (1999-2010) linked to the National Death Index (2011), to determine how nutrition data may advance cardiovascular mortality prediction beyond traditional risk factors (age, sex, race/ethnicity, systolic blood pressure, blood pressure treatment, tobacco smoking, diabetes status).
We will compare our deep learning approach to four standard metrics of nutritional quality: the Healthy Eating Index (HEI-2015), Alternative Healthy Eating Index (AHEI-2010), the Mediterranean Diet Score (MDS), and DASH Diet Score.
Authors: Anirudh Jain anirudhj@stanford.edu, Pranav Samir Rajpurkar pranavsr@stanford.edu, Sanjay Basu basus@stanford.edu