Can we map diets as networks? Food combinations linked to mortality, cardiovascular disease and life expectancy

Imagine a typical breakfast: coffee may appear alongside eggs, toast and butter, while fruit, yogurt and whole grains may form another combination. We do not eat foods in isolation. Instead, our diets are made up of interconnected food combinations. Traditional dietary pattern research has helped us understand overall eating habits, but it does not always show which food groups remain directly connected after the rest of the diet is taken into account.

When I began this study, I wanted to know whether we could map relationships among foods in the same way we map a social network. More importantly, could these dietary networks help us understand the relationships between diet and long-term health outcomes?

To explore these questions, my colleagues and I used machine learning-based network analysis to examine dietary data from Canadian adults and investigate how the identified dietary networks were associated with mortality, cardiovascular disease and life expectancy.

Our research is published in The Journal of Nutrition.

Seeing diet as a network

We analyzed nationally representative Canadian nutrition survey data linked to health administrative databases. To understand how foods are connected, we treated each food group as a node in a network and represented conditional associations between food groups as edges connecting the nodes.

A conditional association means that two food groups remain statistically related after accounting for all other food groups in the diet. Only these relationships were retained as connections in the network.

We used a semiparametric Gaussian copula graphical model to construct the dietary networks. This method is particularly suitable for dietary intake data because food intakes are often highly skewed, and many participants may report no intake of certain foods on a given day.

We then used the Louvain community detection algorithm to group more closely connected food groups into distinct dietary communities. These communities represent dietary patterns composed of food groups that are more closely connected to one another.

Turning dietary communities into individual scores

Identifying dietary networks and communities was only the first step. The networks describe dietary structures at the population level, but to examine how these patterns relate to long-term health outcomes, we also needed to measure how closely each participant followed the pattern represented by each dietary community.

We used eigenvector centrality to measure the importance of each food group within a community. A food group received a higher weight if it was connected not only to many other foods, but also to other influential foods in the community. We defined the food group with the highest weight as the central food group and also considered whether each of the other food groups was positively or negatively related to it.

We then combined each participant’s standardized food intake with the corresponding weight and direction and summed these values within each community to generate an individual community score. A higher score indicated that a participant’s diet more closely reflected the dietary pattern represented by that community.

This novel scoring method preserved both the importance of each food group within the network and the direction of its contribution. It therefore prevented participants who mainly consumed foods positively related to the central food group from receiving scores similar to those who mainly consumed foods related in the opposite direction.

Three dietary communities showed different associations with health

We identified three major dietary communities. The first was a vegetable-rich community, which included several types of vegetables, legumes and soy, soup, pasta and rice, and vegetable oils; fast food was negatively associated with the community’s central foods. The second was a high-sugary beverage and low-fruit community, centered on sugary beverages.

Refined grains and salty snacks were positively associated with the central food, whereas water, whole grains, whole fruits, yogurt, and nuts and seeds were associated in the opposite direction. The third was a high-fat breakfast community, which included solid fats, coffee, sugar, processed meat, eggs, pancakes and waffles; tea was the only food group negatively associated with the central food.

These communities showed different associations with health outcomes. Comparing the 90th with the 10th percentile of community scores, higher vegetable-rich community scores were associated with 51% lower all-cause mortality overall, including 59% lower mortality among females and 48% lower mortality among males. Higher scores were also associated with 45% lower cardiovascular disease risk overall and estimated increases in average life expectancy of 8.3 years for females and 6.1 years for males.

In contrast, higher scores on the high-sugary beverage and low-fruit community were associated with 31% higher mortality overall and 39% higher mortality among males; the association among females was not statistically significant. The high-fat breakfast community was not significantly associated with all-cause mortality or cardiovascular disease risk.

What these findings may mean for dietary guidance

The broader dietary takeaway is that a pattern centered on a variety of vegetables and other plant foods was associated with more favorable long-term health outcomes, whereas a pattern characterized by high sugary beverage intake and low fruit intake was associated with less favorable outcomes, particularly among males.

These findings reflect observational associations and do not establish that any particular dietary pattern directly reduces mortality risk or extends life. Nevertheless, our novel scoring method shows how complex food-group connections can be translated into individual-level measures and linked to long-term health outcomes.

With further validation in other populations, dietary network analysis may help inform dietary guidance and chronic disease prevention strategies that are more food-specific, better tailored to different populations, and attentive to sex differences.

This story is part of Science X Dialog, where researchers can report findings from their published research articles. Visit this page for information about Science X Dialog and how to participate.

Who’s behind this story?

Sadie Harley

BSc Life Sciences & Ecology. Microbiology lab background with pharmaceutical news experience in oil, gas, and renewable industries.

Full profile →

Advertisements

Robert Egan

Bachelor’s in mathematical biology, Master’s in creative writing. Well-traveled with unique perspectives on science and language.

Full profile →

Yifei Wang is a Ph.D. student in Human Nutrition at the University of British Columbia, where she conducts research in the Nutritional Epidemiology and Big Data Analytics (NEBA) Laboratory. Her research combines nutritional epidemiology, machine learning, causal inference and population data science to investigate dietary patterns and chronic disease outcomes using large, nationally representative datasets.

Mahsa Jessri is an Associate Professor of Food, Nutrition and Health at the University of British Columbia and holds the Canada Research Chair (Tier 2) in Nutritional Epidemiology for Population Health. She leads the Nutritional Epidemiology and Big Data Analytics (NEBA) Laboratory. Her research uses linked population data, causal methods and simulation tools to evaluate nutrition policies, dietary patterns and food environments, with a focus on health equity and population health outcomes. Her work aims to translate population-level evidence to inform nutrition policies and chronic