Beyond the label: a methodological and geographic synthesis of ultra-processed food health outcomes

Abstract

The concept of ultra-processed food (UPF) has shifted nutritional epidemiology from single nutrients towards food processing. Higher UPF intake is associated with adverse health outcomes across a large and broadly consistent body of evidence, yet the magnitude and specificity of associations vary across major prospective cohorts: some report linear associations with all-cause mortality, whereas others find risk concentrated in particular subgroups. This critical narrative review synthesises evidence from seminal cohorts (NutriNet-Santé, Nurses’ Health Study (NHS)/Health Professionals Follow-up Study (HPFS), EPIC, UK Biobank) to examine methodological and contextual factors that may contribute to this heterogeneity. We propose three interacting, hypothesis-generating dimensions. (1) The proposed Metric Trap: weight-based metrics (g/day) capture non-nutritive items such as artificially sweetened beverages, whereas energy-based metrics (% kcal) may under-capture them – a proposed ‘Zero-Calorie Paradox’. (2) The proposed Geographic Paradox: identical food categories may exert divergent effects depending on baseline diet quality; for example, ultra-processed whole grains appear neutral – or are excluded from the exposure a priori – in some US analyses, while industrial breads show neutral-to-adverse associations in parts of Europe. (3) Mechanistic Hierarchy: UPF risk may be cumulative, potentially involving nutrient-poor profiles, food-matrix degradation, and neo-formed contaminants. We argue that UPF is not monolithic, while emphasising that this heterogeneity does not negate the broadly consistent overall associations. We propose a Risk-Stratified Typology that moves beyond binary ‘High vs. Low’ comparisons towards scenario-specific definitions (toxicological, nutritional, volumetric). These frameworks are interpretive hypotheses requiring formal empirical validation.