Ultra-Processed Food Intake and Risk of Type 2 Diabetes Mellitus: A Dose-Response Meta-Analysis of Prospective Studies (Diabetes Metab J 2025;49:1064-74)

Article information

Diabetes Metab J. 2026;50(1):194-195

Publication date (electronic) : 2025 December 15

doi : https://doi.org/10.4093/dmj.2025.0925

Lirong Hu¹, Aiji Chen¹, Gang Tian^,2

¹School of Stomatology, Southwest Medical University, Luzhou, China

²Department of Laboratory Medicine, The Fourth Affiliated Hospital of Southwest Medical University, Sichuan, China

Corresponding author: Gang Tian https://orcid.org/0000-0002-4116-3215 Department of Laboratory Medicine, The Fourth Affiliated Hospital of Southwest Medical University, No. 168, Huantianfu New Area Expressway, Meishan Tianfu New Area, Meishan (Sichuan) 620564, China E-mail: tiangang@swmu.edu.cn

Kim et al. [1] synthesized 12 prospective cohorts and reported a positive association between ultra-processed food (UPF) intake and incident type 2 diabetes mellitus, providing valuable evidence on the potential link between UPF consumption and type 2 diabetes mellitus. This offers a basis for future dietary guidance, while several methodological issues and data inconsistencies deserve careful discussion to ensure the robustness and generalizability of the findings.

First, the diagram indicates ‘records identified, n=569,’ while listing 483 records from PubMed, 685 from Web of Science, and 1,080 from Embase (totaling 2,248). If 2,248 represents the pre-deduplication yield, removing 89 duplicates would result in 2,159 unique records, of which only 405 were screened—suggesting 1,754 citations were excluded without explanation. Conversely, if 569 is the actual number identified after merging, it does not align with the database-specific totals provided. This discrepancy in numbers obscures the screening denominator, hinders independent replication of the search, and raises concerns about potential pre-screening filters. It is recommended to publish corrected figures, a deduplicated bibliography, and detailed search strings in accordance with Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 guidelines to ensure reproducibility [2].

Second, the perceived ‘nonlinearity’ observed in the relationship between UPF consumption and diabetes risk may stem from sparse data points at extreme intake levels. Specifically, Fig. 2D in original article shows that although all studies covered a range of >300 g/day, the data points were sparse (each study had only 1 to 2 extreme value intervals), which could affect the stability of the curve. The amalgamation of diverse exposure units (e.g., %, g/day, servings/day) across multiple cohorts further complicates interpretation. To enhance the reliability of the analysis, it is suggested to define spline knots a priori, confine the interpretation to the range of observed values, and employ a one-stage random-effects spline model that accounts for within-study covariance. Harmonizing exposure metrics (e.g., g/1,000 kcal or % energy) is also recommended to minimize the influence of unit-specific curvature. These methodological adjustments offer a more robust evaluation of nonlinearity and help mitigate the risk of misinterpreting threshold effects [3].

Third, it is likely that exposure misclassification occurred since the majority of cohorts relied on food-frequency questionnaires (FFQs) that were not tailored to capture processing levels. The translation of FFQ line-items into NOVA classes involves making assumptions and seeking expert opinions, which may differ, especially for breads, cereals, and mixed dishes [4]. To address residual confounding arising from dietary quantity and measurement inaccuracies, we also propose aligning exposure measurements on a standardized scale (e.g., grams per 1,000 kcal or as a percentage of energy) and implementing regression calibration methods across all cohorts [5]. Besides, the substantial heterogeneity (I²=73.3%) coupled with moderate study numbers suggests that Hartung-Knapp-Sidik-Jonkman adjustment would provide more conservative confidence intervals than the employed DerSimonian-Laird method [6]. The body mass index-mediated pathway, evidenced by attenuated but persistent associations after adjustment (relative risk, 1.48 to 1.29), aligns with an inpatient randomized controlled feeding trial demonstrating UPF-induced hyperphagia [7]. Heterogeneous outcome ascertainment across cohorts, ranging from self-report to biochemical verification, warrants stratified sensitivity analyses to assess potential misclassification bias [8].

In conclusion, this meta-analysis emphasizes a credible and policy-relevant connection between UPF exposure and the risk of diabetes, offering a significant indication for public health guidance. Improving exposure harmonization and validation, rectifying flow-diagram inconsistencies, and employing robust meta-analytic estimators will enhance the robustness and generalizability of the conclusions.