A recent study published in Nature Food used advanced machine learning techniques to analyze over 50,000 food products from leading US grocery store websites. This research led to the development of the GroceryDB database, which serves as a valuable tool for both consumers and public health initiatives. It provides insights into the degree of food processing in grocery stores, helping shoppers make healthier decisions while addressing concerns about ultra-processed foods (UPF).
Ultra-processed foods (UPF) have been identified as a significant contributor to health problems worldwide, particularly in developed countries, where UPF accounts for up to 60% of total calorie intake. Grocery stores are the primary outlet through which these foods reach consumers, raising concerns about how to effectively quantify the extent of food processing and find alternatives to reduce UPF consumption.
Quantifying food processing is challenging, as food labels often present mixed or unclear information, creating ambiguity. To address this, scientists have called for a more objective definition based on biological mechanisms, while AI technologies like machine learning are being increasingly employed to advance nutrition security and better understand the impact of food processing on health.
The study involved compiling publicly available data from major grocery stores in the US, including Walmart, Target, and Whole Foods. Researchers used these data to classify food items, ensuring consistency by aligning the classification systems used by each store. Nutrient concentrations were standardized, and FoodProX, a random forest classifier, was employed to assess the degree of food processing for each product. This system, referred to as the Food Processing Score (FPro), analyzes the changes in nutrient quantities that occur due to food processing.
Extensive testing validated the stability of the FPro score. The final score reflected the probability of observing nutrient patterns typical of unprocessed food versus those seen in UPF. Additionally, variations in the price per calorie across different levels of food processing were analyzed using robust linear models.
The analysis of over 50,000 food products revealed some interesting patterns. While the majority of grocery store inventory consisted of highly processed or ultra-processed foods (high FPro), the low FPro foods, representing minimally processed items, made up a much smaller proportion of the available options. However, consumers tended to purchase a higher percentage of low FPro items, revealing a disconnect between store inventory and consumer preferences.
The study also highlighted some store-specific differences. For example, Whole Foods offered fewer ultra-processed foods compared to Target, which had a higher proportion of high FPro products. Low FPro foods were more limited in categories like jerky, popcorn, biscuits, mac and cheese, chips, and bread, but there was more consumer choice in other categories such as cereals, pasta noodles, milk, and snack bars. The FPro distribution in GroceryDB closely mirrored that of the USDA’s Food and Nutrient Database for Dietary Studies (FNDDS).
A notable finding from the study was the relationship between the price and the degree of food processing. As the level of food processing (FPro) increased, the price per calorie decreased. On average, a 10% increase in FPro led to an 8.7% reduction in price per calorie across all food categories. This trend was especially evident in processed foods, which were typically cheaper per calorie than their minimally processed counterparts. In the milk and milk substitutes category, however, the relationship showed an upward trend, suggesting that less processed options could become more affordable over time.
The study also examined the differences between grocery stores for the same food categories. For example, cereals sold at Whole Foods contained fewer artificial flavors, less sugar, and fewer added vitamins compared to those offered at Walmart and Target. The range of FPro scores also varied across food items, with some categories like pizza, popcorn, and mac and cheese being highly processed at all stores, while others, like cookies and biscuits, had a wider variety of FPro scores at Whole Foods.
An ingredient-based FPro (IgFPro) score was calculated to show how individual ingredients contribute to the overall level of processing in food products. The analysis demonstrated that more complex ingredient lists typically led to higher levels of processing.
This groundbreaking research uses machine learning to model the chemical complexity of food items from major supermarkets. The creation of GroceryDB and the development of FPro provide a data-driven approach for consumers to identify healthier, less processed alternatives across a variety of food categories. By harnessing these tools, consumers can make more informed decisions about their food choices, ultimately promoting better nutrition and public health. The insights gained from this study could guide future initiatives aimed at reducing UPF consumption and improving the overall health of populations.
You Might Be Interested In:
-
Gut Microbes and the Body Work Together to Regulate Fat Metabolism
-
$20 Million Grant to Investigate Viruses’ Impact on Human Health
-
Study Links Fiber Consumption to Epigenetic Changes with Anti-Cancer Effects
