Dietary Patterns and the Social Development Index in Mexico City: A Machine-Learning Approach

Background: Dietary composition reflects the interplay of individual and contextual factors, yet its relationship with socioeconomic–territorial conditions remains insufficiently characterized in large Latin American cities. Methods: We identified empirical dietary patterns among 3439 adults from the cohort in Mexico City and examined their association with the Social Development Index (SDI), a composite area-level measure of socioeconomic–territorial development. Energy-adjusted food-group intakes were analyzed using k-means clustering, and supervised machine-learning models with SHAP-based interpretation were used to characterize pattern membership. Results: Three dietary patterns—Fresh, Mixed, and Refined—were identified along a continuous dietary gradient primarily differentiated by the relative contribution of fruits and vegetables. SHAP analysis showed that the Fresh pattern was characterized mainly by female sex and older age, the Refined pattern by greater contributions from physical activity, cardiometabolic markers, male sex, and lower SDI values, whereas the Mixed pattern was distinguished primarily by cardiometabolic variables, alcohol intake, and durable-goods ownership. Dietary pattern membership was significantly associated with SDI stratum (χ2=40.77, p<0.001), although the effect size was small (Cramér’s V = 0.078). The proportion of participants classified in the Fresh pattern increased from 35.7% in the lowest SDI stratum to 51.9% in the highest, whereas the Refined pattern decreased from 44.6% to 34.0%. Supervised models showed limited discriminatory performance (balanced accuracy ≈ 0.40), indicating substantial overlap among patterns despite their distinct multivariable profiles. Conclusions: These findings indicate that dietary variation in this urban population is better represented as a continuum than as sharply separated dietary phenotypes.