Consumption of ultraprocessed foods through item response theory

This is a psychometric study aimed at proposing a scale for estimating the consumption of ultraprocessed foods using item response theory.

Food consumption data from a representative sample of 2515 adolescent individuals aged 18 and 19 from the third phase of the Ribeirão Preto, Pelotas, and São Luís Brazilian Birth Cohorts Consortium were used. The instrument used was a validated food frequency questionnaire. The selection of ultraprocessed foods items occurred through exploratory factor analysis. Items with factor loadings > 0.30 and commonality > 0.20 were applied in Samejima's graded response model and had unidimensionality (variance = 28.0%) as well as good quality of fit (comparative fit index and Tucker-Lewis index > 0.90 and root mean square error of approximation < 0.08). Kruskal-Wallis and Wilcoxon tests were used to assess latent trait levels in thirds (low, moderate, and high), according to food consumption variables.

The information provided by the test covers the entire latent trait continuum, reinforcing its ability to estimate ultraprocessed foods consumption. Foods classified as in natura had a decline in consumption as the level of ultraprocessed foods consumption increased. Ultraprocessed foods were directly proportional to the latent trait (P < 0.05). The highest level of latent trait concentrated the highest median of total energy intake (P < 0.05).

The proposed ultraprocessed foods consumption scale presents an estimation of consistency and is a potential brief tool for use in assessing the consumption of this food group.