Machine learning links up to 64 percent of Salmonella cases to chicken and vegetables

A study by the CDC, FDA and USDA used whole-genome sequencing (WGS) and machine learning to identify chicken and vegetables as leading sources of Salmonella enterica infections in the United States.

The research, published in the April 2025 issue of Emerging Infectious Diseases, focused on sporadic cases, which account for 95 percent of the Centers for Disease Control and Prevention’s estimated 1.35 million annual Salmonella illnesses.

Since WGS became routine in 2019, this approach could improve outbreak tracking and inform food safety policies. The findings highlight chicken and vegetables as key Salmonella sources, underscoring the need for stronger contamination controls.

Model reveals food sources
Researchers trained a random forest model using WGS data from 18,661 Salmonella isolates sourced from single food categories: chicken at 31.3 percent, vegetables at 12.8 percent, turkey at 12.3 percent, and pork at 10.9 percent. Using 7,360 genetic loci, the model achieved 91 percent accuracy overall, including 97 percent accuracy for chicken isolates.

When applied to 6,470 human isolates collected by the CDC’s FoodNet program from 2014 to 2017, the model linked 34 percent of human cases to chicken and 30 percent to vegetables. At a higher confidence threshold of 50 percent, these estimates shifted to 46 percent and 27 percent, respectively.

Serotype patterns emerge
The study associated specific Salmonella serotypes with food sources: chicken with Enteritidis, Typhimurium, Heidelberg, and Infantis; vegetables with Javiana and Newport; and pork with 4,[5],12:i:-, a serotype named for its antigenic markers, which resembles Typhimurium and is frequently linked to swine.

These findings differ from previous outbreak-based estimates, which attributed only 17 percent of Salmonella cases to poultry. Researchers noted that sporadic and outbreak infections may carry different risk patterns, and contamination points in the food supply chain could influence these variations.

Because of the overrepresentation of chicken in the training data, adjustments were made to the model. Additionally, because the study focused solely on foodborne sources, it did not account for environmental or animal transmission, though food accounts for an estimated 66 percent of Salmonella cases. Incorporating water or animal-derived isolates in future models could enhance accuracy.

The full study can be found here.

(To sign up for a free subscription to Food Safety News, click here.)