Food-Consumption Profiles and Context-Specific Associations Among University Students: An Exploratory Multicountry Person-Centred Clustering and CHAID Analysis

Background: University students often report suboptimal dietary practices, but overall diet-quality scores may conceal differences in food-group consumption. This study explored whether distinguishable food-consumption profiles could be identified among university students from Chile, Mexico, Spain, and Italy, and examined, exploratorily, the characteristics associated with profile membership. Methods: In this cross-sectional study, 686 university students completed a dietary assessment using nine component scores of the Healthy Eating Index adapted for the Spanish population (IASE). Profiles were derived by Ward’s hierarchical clustering with k-means refinement and assessed for split-half stability; Exhaustive CHAID explored characteristics associated with profile membership. Results: Internal validity indices favoured no candidate solution (silhouette 0.113–0.126 across two- to five-cluster solutions), and split-half stability was limited throughout, including for the retained four-cluster solution (mean Adjusted Rand Index = 0.30). Four profiles were nonetheless retained: Healthy and Prudent (24.8%), Low Plant-Food (13.0%), Mixed Dietary (36.6%), and Protein-Oriented (25.7%); all occurred in every country. Country was the first CHAID split, followed by living arrangement (Chile/Mexico), year of study (Spain), and post-pandemic PA change (Italy). The model did not outperform the majority-class baseline; the segmentation is therefore exploratory, not predictive. Conclusions: A four-cluster configuration could be derived in each setting, but validity and stability indices indicate it is not robust; country/academic-programme—confounded with institution—was the only characteristic with a non-negligible, though modest, association with profile membership. Findings are hypothesis-generating and require replication in larger, representative, longitudinal samples before profiles are treated as an established typology.