Yujin Kim; Hannah Oh

doi:10.4093/dmj.2025.1104

Articles

Page Path: HOME > Diabetes Metab J > Volume 50(1); 2026 > Article

Response 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): Yujin Kim¹, Hannah Oh^1,2; Diabetes & Metabolism Journal 2026;50(1):208-210.
DOI: https://doi.org/10.4093/dmj.2025.1104
Published online: January 1, 2026

741 Views
28 Download

¹Interdisciplinary Program in Precision Public Health, Department of Public Health Sciences, Graduate School of Korea University, Seoul, Korea

²Department of Health Policy and Management, College of Health Sciences, Korea University, Seoul, Korea

Corresponding author: Hannah Oh

Department of Health Policy and Management, College of Health Sciences, Korea University, Hana Science Building B358, 145 Anam-ro, Seongbuk-gu, Seoul 02841, Korea Email: hannahoh@korea.ac.kr

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

prev next

Full Article

Download PDF

NOTES
REFERENCES
About this article

See the letter "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)" on page 194.

We appreciate Dr. Lirong Hu and colleagues for their thoughtful comments on our recently published article, “Ultra-processed food intake and risk of type 2 diabetes mellitus: a dose-response meta-analysis of prospective studies” [1]. We would also like to thank the editor for the opportunity to respond and further discuss our work.

First, we apologize for the typographical errors in the original diagram in Supplementary Fig. 1. While the total number of references identified was correctly reported (n=569), the breakdown by database contained errors and should be corrected to ‘134 from PubMed, 167 from Web of Science, and 268 from Embase,’ totaling 569 references. Among these 569 references, 164 duplicates were removed, leaving 405 references for title and abstract screening. The original numbers in the article represented references identified before applying search terms that restricted the search of keywords to titles and abstracts. We have processed a corrigendum to make the necessary corrections to the article. The detailed search terms were provided in Supplementary Table 1 of the original article, in accordance with Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 guidelines [2].

We agree that the limited data for extreme levels of ultra-processed food (UPF) intake may have produced spurious nonlinearity. As indicated by the wider confidence intervals at extreme intake levels, these data should be interpreted with caution. In our analysis, we used a priori defined spline knot placement at the 10th, 50th, and 90th percentiles of UPF intake, and the interpretation of the dose-response curve was restricted to the range of observed values. The observed mean (or median) intake in the lowest and highest UPF categories ranged from 107.1 to 579.5 g/day in CARTaGENE, 240 to 677 g/day in European Prospective Investigation into Cancer and Nutrition (EPIC), 166 to 758 g/day in The Brazilian Longitudinal Study of Adult Health (ELSA-Brazil), and 140.0 to 481.6 g/day in Seguimiento Universidad de Navarra (SUN)-Project. In EPIC and ELSA-Brazil, the population mean (or median) intake level was also above 300 g/day (413 g/day for men and 326 g/day for women in EPIC; 372 g/day in ELSA-Brazil). Thus, several cohorts contributed substantial information beyond 300 g/d of UPF intake, the range in which the evidence of nonlinearity was observed.

In our analysis, we used two-stage random-effects dose-response models. A previous study suggested that one-stage analysis may be advantageous when studies contribute few exposure categories or when between-study heterogeneity is substantial [3]. Given the exposure categories and low-to-moderate heterogeneity in the g/day and serving/day units, the two-stage approach appeared appropriate in our analysis. The overall finding of a positive dose-response association, with no upper limit of risk elevation, was consistently observed with both one-stage and two-stage random-effects models. However, the evidence of a potential threshold effect was observed in the two-stage models only (P-nonlinearity=0.001) and did not persist in the one-stage models (P-nonlinearity=0.354), suggesting that the observed nonlinear association may be sensitive to between-study variability. Therefore, further studies are needed to confirm our findings.

In our study, we also conducted a dose-response meta-analysis separately for each UPF unit. The rationale of this approach was to investigate the unit-specific dose-response relationships between UPF intake and diabetes risk. Each unit reflects different aspects of the relationship. For example, weight ratio (the percentage of g/day) reflects the composition of non-nutritional factors (e.g., food additives) that do not contribute to energy intake, while the absolute g/day reflects the absolute quantity of consumption and the serving/day accounts for portion size. We provided additional insight into these relationships by quantifying the risks associated with specific levels and units of UPF intake.

As we previously acknowledged, most studies included in our meta-analysis used semi-quantitative food frequency questionnaire (FFQ) dietary assessment methods, which often lacked sufficient information (e.g., degree and purpose of processing) to classify UPFs according to the Nova definition. This also highlights an important limitation repeatedly acknowledged in many UPF observational studies to date. To advance our understanding of the effects of UPFs on various health outcomes, more valid dietary assessment tools specifically designed to capture UPF intake are required. Although the FFQ has advantages over single-day 24-hour recall and dietary records in assessing individuals’ long-term average intake, the validity of its estimates depends on the comprehensiveness and appropriateness of food items included in the questionnaire. Because the same food items may exist in both UPF and non-UPF forms, the FFQ should also include more detailed descriptions and different forms of food items. For example, plain Greek yogurt without food additives is not considered a UPF, whereas commercial fruit-flavored yogurt containing fruit extract and other food additives is classified as a UPF. Therefore, both UPF and non-UPF forms of yogurt should be included in the FFQ food list to distinguish between minimally processed and ultra-processed products. In South Korea, white rice can be consumed either as a minimally processed food (e.g., home-cooked rice) or as a UPF (e.g., ready-to-eat microwavable rice), and thus both forms should be included in the FFQ to accurately estimate UPF intake.

Hu and colleagues also suggested several additional analyses, such as the Hartung-Knapp-Sidik-Jonkman (HKSJ) adjustment and stratification by outcome ascertainment. The pooled relative risk (RR) and I² estimates were nearly identical when we conducted additional analysis applying the HKSJ adjustment (RR, 1.48; 95% confidence interval [CI], 1.35 to 1.62; I²= 73.3%) compared with our original estimates using the DerSimonian Laird (DL) method (RR, 1.48; 95% CI, 1.36 to 1.61; I²= 73.3%). A previous simulation study has shown that the DL method may yield inflated error rates when the number of studies is small (<10) or the study-size distribution is extremely unbalanced [4]. However, as the number of studies included in the meta-analysis increases, the difference between the DL and HKSJ methods becomes smaller [4], yielding similar results. In our meta-analysis of 12 prospective cohorts, the study-size distribution was not skewed (Supplementary Table 2 of our original article), justifying our method. In terms of outcome ascertainment, there was some variation in methods among the studies. Four cohorts (Australian Longitudinal Study on Women’s Health [ALSWH], Nurses’ Health Study [NHS], Nurses’ Health Study II [NHSII], Health Professionals Follow Up Study [HPFS]) relied solely on self-reported data, which may be subject to measurement errors, whereas three other cohorts (Korean Genome and Epidemiology Study [KoGES] Ansan-Ansung, ELSA-Brazil, Lifelines) validated the outcome data using laboratory measurements and five cohorts (CARTaGENE, EPIC, SUN-project, NutriNet Santé, UK Biobank) validated against medical records. However, in our subgroup analysis, we identified no significant between-group heterogeneity by outcome ascertainment (P-heterogeneity=0.797; RR, 1.48 [95% CI, 1.25 to 1.76] with self-reported outcome data vs. 1.44 [95% CI, 1.33 to 1.56] with outcome ascertainment using biochemical or medical records) (Fig. 1).

In conclusion, the meta-evidence supports that higher UPF intake is associated with higher risk of type 2 diabetes mellitus in a monotonic dose-response fashion. Although a growing number of studies have investigated associations between UPF intake and type 2 diabetes mellitus, questions still remain, underscoring the need for further research.

NOTES

CONFLICTS OF INTEREST

No potential conflict of interest relevant to this article was reported.

Fig. 1

Subgroup analyses by outcome ascertainment for the association between combined ultra-processed food (UPF) intake (highest vs. lowest categories) and type 2 diabetes mellitus risk. This figure shows the results of subgroup analysis by outcome ascertainment (P-heterogeneity=0.434) for the association between combined UPF intake (highest vs. lowest categories) and type 2 diabetes mellitus risk. The black diamonds and horizontal lines from forest plots represent the study-specific relative risk (RR) and 95% confidence interval (CI). The weight of each of the studies is represented by the size of the gray square. The overall effect estimate and corresponding 95% CI are represented by a hollow diamond. P-heterogeneity was calculated by Cochran’s Q test. ALSWH, Australian Longitudinal Study on Women’s Health; NHS, Nurses’ Health Study; NHSII, Nurses’ Health Study II; HPFS, Health Professionals Follow Up Study; DL, DerSimonian-Laird method; KoGES, Korean Genome and Epidemiology Study; EPIC, European Prospective Investigation into Cancer and Nutrition; ELSA-Brazil, The Brazilian Longitudinal Study of Adult Health; SUN, Seguimiento Universidad de Navarra.

REFERENCES

1. Kim Y, Cho Y, Kim JE, Lee DH, Oh H. 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 PubMed PMC PDF
2. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. J Clin Epidemiol 2021;134:178-89.PubMed
3. Crippa A, Discacciati A, Bottai M, Spiegelman D, Orsini N. One-stage dose-response meta-analysis for aggregated data. Stat Methods Med Res 2019;28:1579-96.Article PubMed PDF
4. IntHout J, Ioannidis JP, Borm GF. The Hartung-Knapp-Sidik-Jonkman method for random effects meta-analysis is straightforward and considerably outperforms the standard DerSimonian-Laird method. BMC Med Res Methodol 2014;14:25.Article PubMed PMC PDF

Figure & Data

References

Citations

Citations to this article as recorded by

PubReader
ePub Link

Cite this Article

Cite this Article: export Copy Download Format; Close

Download Citation

Download a citation file in RIS format that can be imported by all major citation management software, including EndNote, ProCite, RefWorks, and Reference Manager.

Format:

RIS — For EndNote, ProCite, RefWorks, and most other reference management software
BibTeX — For JabRef, BibDesk, and other BibTeX-specific software

Include:

Citation for the content below
Citation and abstract for the content below

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)

Diabetes Metab J. 2026;50(1):208-210. Published online January 1, 2026

DOI: https://doi.org/10.4093/dmj.2025.1104

XML Download

Figure

Related articles

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)

Fig. 1 Subgroup analyses by outcome ascertainment for the association between combined ultra-processed food (UPF) intake (highest vs. lowest categories) and type 2 diabetes mellitus risk. This figure shows the results of subgroup analysis by outcome ascertainment (P-heterogeneity=0.434) for the association between combined UPF intake (highest vs. lowest categories) and type 2 diabetes mellitus risk. The black diamonds and horizontal lines from forest plots represent the study-specific relative risk (RR) and 95% confidence interval (CI). The weight of each of the studies is represented by the size of the gray square. The overall effect estimate and corresponding 95% CI are represented by a hollow diamond. P-heterogeneity was calculated by Cochran’s Q test. ALSWH, Australian Longitudinal Study on Women’s Health; NHS, Nurses’ Health Study; NHSII, Nurses’ Health Study II; HPFS, Health Professionals Follow Up Study; DL, DerSimonian-Laird method; KoGES, Korean Genome and Epidemiology Study; EPIC, European Prospective Investigation into Cancer and Nutrition; ELSA-Brazil, The Brazilian Longitudinal Study of Adult Health; SUN, Seguimiento Universidad de Navarra.

Fig. 1

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)

About this article

Kim Y, Oh H. 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). Diabetes Metab J. 2026;50(1):208-210.

DOI: https://doi.org/10.4093/dmj.2025.1104.

Download citation ↓