(P04-083-20) An Unsupervised Machine-Learning Approach to Characterize Ultra-Processed Food Consumers: Results from the NHANES 2015-2015 Survey Cycle

Objectives: The consumption of ultra-processed food (UPF) has been suggested to be associated with adverse health outcomes.  Understanding factors that affect the consumption of UPF is relevant from the public health perspective in order to develop ad hoc public health strategies facing UPF consumption. The present study aimed at assessing the contribution of demographic and socio-economic data in characterizing UPF consumers using the NHANES 2015-2016 survey cycle data.

Methods: Demographic and socio-economic data reported in the interview by NHANES 2015-2016 adult participants (age > 18 years) were considered in the analysis. All the foods reported in the interview were classified to be or not UPF according to the NOVA classification. The consumption of UPF was assessed using dietary data for day 1 interview and it was reported as percentage of total energy intake. Subjects were classified by the percentage of UPF energy intake (less or more than 50% of total energy intake). A Factor Analysis for Mixed Data (FAMD) was run on subjects with an UPF intake that was above the 50% of total energy intake using the demographic and socio-economic data as explanatory variables.

Results: Analyses were performed on 5266 survey participants that were found to be UPF consumers; of these, 2669 had an UPF intake above the 50% of the total energy intake. The results of the FAMD showed that the first three latent dimensions identified explained the 67.5% of overall data space. The main contributors of the first latent dimension were the family income (32.14%), the ratio of family income to poverty (31.67%), and the annual household income (27.36%). The factors that contributed the most to identify the second latent dimension were the number of people in the household (37.57%), the number of children aged 6-17 years old and of preschoolers living in the household (24.84% and 17.52%, respectively).

Conclusions: Among the factors considered, those related to the socio-economic characteristics were found to contribute the most in characterizing UPF consumers. Such results suggest that UPF consumption could be affected by socio-economic status that should be taken into account in the set-up of public health intervention strategies facing UPF consumption.

Funding Sources: University of Padova