Blood-based metabolic signature outperforms standard method for predicting diet, disease risk

Dec. 14, 2022
The study found that machine learning techniques improved diet prediction by 10-20%.

A research team led by a Michigan Medicine cardiologist have found a method using molecular profiling and machine learning to develop blood based dietary signatures that more accurately predict both diet and the risk of cardiovascular disease and type 2 diabetes. The results are published in European Heart Journal.

Researchers followed more than 2,200 white and Black adults in the Coronary Artery Risk Development in Young Adults study, using blood samples and food surveys to determine metabolite signatures of diet and subsequent disease risk over 25 years. Through a machine learning model, investigators were able to create a blood-based dietary signature that more accurately predicts a person’s entire diet over 19 food groups by 10-20%. 

Additionally, the blood-based signature often outperformed the healthy eating index, a standard measure of diet quality, for identifying who is more likely to develop both diabetes and cardiovascular disease based on each food group. For example, when the food frequency questionnaire indicated an 18% increase in the risk of diabetes for a person eating red meat, the blood-based signature found a 55% increased risk. 

Read the report.