Study Finds Machine Learning Techniques Improved Diet Prediction by 10-20% – ScienceDaily

When it comes to studying food and nutrition, it’s difficult to know what people eat — let alone their risk of disease caused by what they eat.

Doctors and researchers typically ask people to fill out a lengthy food frequency questionnaire that estimates caloric intake, food groups, and nutrients. That depends on a person’s memory and may not provide the most accurate picture.

However, a research team led by a Michigan Medicine cardiologist has found a method that uses molecular profiling and machine learning to develop blood-based nutritional signatures that predict both diet and the risk of cardiovascular disease and type 2 diabetes predict more accurately. The results will be published in European Heart Journal.

“Diet is not one-dimensional; it’s ever-changing, and the way we traditionally assess it isn’t perfect,” said senior author Venkatesh Murthy, MD, Ph.D., cardiologist at the University of Michigan Health Frankel Cardiovascular Center and associate professor of cardiology at the UM Medical School.

“We need tools that are more reliable and precise, while being easy for everyone to use. Using metabolite signatures and data science, we can better understand how much people actually eat and what their risks are for the cardiovascular disease that affects millions of Americans,” Murthy said.

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

READ :  UTC Scientist Hong Qin A Round One Winner In Global Research Challenge

In addition, the blood-based signature often outperformed the Healthy Eating Index, a standard measure of diet quality, to determine who is more likely to develop both diabetes and cardiovascular disease based on each food group. For example, if the food frequency questionnaire indicated an 18% increased risk of diabetes in a person who eats red meat, the blood-based signature revealed a 55% increased risk.

“The use of metabolites to understand food stress and nutrition is an expanding area in nutritional science,” said co-author Ravi Shah, MD, cardiologist and associate professor of medicine at Vanderbilt University Medical Center. “In addition to understanding which types of diets are better or worse for our health, the methods here could enable those studying food science to create a metabolic snapshot of diet and nutrition to better understand their effects on health.”

The results follow a $170 million award from the National Institutes of Health to clinics and centers across the country for a Nutrition for Precision Health study for “develop[ing] Algorithms to predict individual responses to foods and dietary routines,” reads a press release.

The blood-based signature technique needs to be tested in prospective, controlled trials using different diets, researchers say. Knowing exactly how well people are sticking to a diet that uses blood-based signatures will yield even stronger results, according to Murthy.

“Nutrition and nutrition research is really difficult,” Murthy said. “We see this as an important step and set of tools to conduct nutritional research with greater precision and efficiency. Ultimately, such work may allow us to better understand optimal nutrition for our patients.”

READ :  EDM Council and National Institute of Standards and Technology (NIST) Collaborate to Launch Ontology Development Infrastructure for the Manufacturing Industry

Additional authors are Lyn M. Steffen, Ph.D., MPH, David R., Jacobs Jr., Ph.D., University of Minnesota School of Public Health, Matthew Nayor, MD, Boston University School of Medicine, Jared P. Reis , Ph.D., National Heart, Lung, and Blood Institute, Norrina B. Allen, Ph.D., Donald Lloyd-Jones, MD, Sc.M., Northwestern University, Katie Meyer, Sc.D., UNC Chapel Hill, Joanne Cole, Ph.D., Massachusetts General Hospital, Paolo Piaggi, Ph.D., University of Pisa, Ramachandran S. Vasan, MD, Boston University Schools of Medicine and Public Health, The Framingham Heart Study, and Clary B Clish , Ph.D., Broad Institute of Harvard and MIT, University of Michigan.

This work was supported by grants from the National Institutes of Health and the American Heart Association.

Leave a Reply

Your email address will not be published. Required fields are marked *