GRAPHICAL ABSTRACT

Highlights

• Ultra-processed food intake is associated with an increased type 2 diabetes risk.

• The association was independent of diet quality, total energy intake, and BMI.

• A threshold effect was observed at intake exceeding 300 g/day.

• Among subgroups, processed meat intake increases diabetes risk.

• Ultra-processed cereals or breads and savory snacks are inversely associated.

INTRODUCTION

According to the Nova classification, there are four different food groups (groups 1–4) classified according to the degree and method of processing [1]. The group 4 represents the most processed food group, called ultra-processed foods (UPFs). UPFs are industrial formulations that are made from fractions of whole foods, often chemically modified and enhanced with food additives such as artificial sweeteners, preservatives, and emulsifiers [2]. UPFs, such as sugar-sweetened beverages, frozen meals, and instant foods, are typically energy-dense; high in fat, sugar, and sodium; and low in fiber [3]. For this reason, UPF intake is likely to result in excessive total energy intake and poor diet quality (or poor nutritional values) [4,5], contributing to increased risks of obesity [6,7] and other chronic diseases [8,9]. High UPF intake may also lead to impaired insulin signaling and thereby increase the risk of type 2 diabetes mellitus (T2DM) [10,11]. Food additives, such as sweeteners and emulsifiers, which are found in most UPFs, can further promote insulin resistance and inflammation by disrupting the gut microbiota-host relationship [12,13].

Previous studies suggested a positive association between UPF intake (highest vs. lowest categories) and the risk of T2DM [10,11,14-21]. However, the magnitude of association varied across the studies. The possible explanations for the variation in association include the differences in the range of UPF intake, UPF unit (percentage of total g/day vs. absolute g/day vs. serving/day), UPF composition (relative contribution of individual UPF items to the total UPF intake), timing of dietary assessment (baseline vs. updated), adjustment variables (e.g., diet quality, total energy intake, body mass index [BMI]), and population characteristics (e.g., geographic region) among the studies. In particular, because studies have used different units of UPF intake, it has been difficult to compare the results across studies and estimate the risk associated with specific units of UPF intake. A recent meta-analysis of seven prospective studies from the United States and European countries reported a 12% higher risk of T2DM risk for every 10% (of g/day) increase in UPF intake [15]. However, little is known about the exact shape and risks associated with other UPF units, such as g/day and serving/day, as well as the association in non-Western populations with different UPF compositions. It is also yet unclear whether the relationship is independent of pathways reflected by diet quality, total energy intake, and BMI.

In this study, we conducted a meta-analysis of prospective studies, including additional studies from both Western (Europe, North America) [14,19] and non-Western populations (South Korea [11], Brazil [10], and Australia [18]) that were not part of the previous meta-analysis [15]. By conducting linear and nonlinear dose-response meta-analyses, we also explored the shape of the relationship and quantified the risks associated with specific levels of UPF intake using three different UPF units (percentage of g/day, absolute g/day, serving/day). Additionally, to identify the sources of heterogeneity in meta-analysis, we conducted subgroup analyses of published studies to investigate whether the associations varied by region, timing of dietary assessment, and adjustment variables.

METHODS

Search strategy

We designed and conducted a systematic review and metaanalysis following the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines [22]. Our study protocol was registered at the international prospective register of systematic reviews (PROSPERO CRD42024503079). Yujin Kim and Yoonkyoung Cho independently conducted the literature search, study selection, and data extraction, and Hannah Oh checked for accuracy. Studies published through January 2024 were identified by searching PubMed, Embase, and Web of Science databases. The detailed search terms used are provided in Supplementary Table 1. The reference lists of previous meta-analyses were also reviewed for additional studies.

Study selection

This meta-analysis was restricted to prospective studies that provided relative risk (RR) estimates (hazard ratios, risk ratios, or odds ratios) along with 95% confidence intervals (CIs) for the association between UPF intake and T2DM risk. Studies were excluded if the exposure (UPF intake) was not categorized by the Nova classification. Articles without full-text or not written in English were also excluded. For our dose-response meta-analysis, we restricted the analysis to studies that reported at least three categories of UPF intake variables with information on the RR, 95% CI, and category-specific number of cases and person-years. A total of 10 publications, covering 12 prospective cohort studies, met the eligibility criteria for inclusion in this study. The detailed study selection procedure is summarized in Supplementary Fig. 1.

Data extraction

For each selected study, we extracted the following data: the last name of the first author, year of publication, name of the cohort, country, total number of study participants, baseline age, average duration of follow-up, unit of UPF intake, dietary assessment method, timing of UPF assessment (baseline vs. updated during the follow-up), number of cases and personyears of all relevant categories necessary for our calculations, range of exposure for each category, covariates included in the fully adjusted model, and RRs and their corresponding 95% CIs by categories or per unit. When the study reported RRs and 95% CIs separately for multiple multivariable-adjusted models (e.g., with vs. without BMI, diet quality, total energy intake), we extracted the estimates from each model separately for each subgroup analysis. For our primary analysis, we used the estimates from the most adjusted model. If the study reported multivariable models both with and without adjustment for BMI, we included the estimate from the model without BMI in the primary analysis because BMI can be a potential mediator. When results from multiple UPF units were reported [11,15], the estimates were separately extracted from each unit for dose-response meta-analyses, and the results from the main study model were used in our meta-analysis that compares the highest vs. lowest categories. In cases where studies reported RRs and 95% CIs for the linear trend (e.g., per 1 g/day increment in UPF intake) but did not report category-specific information (RR, 95% CI, or number of cases and person-years), we contacted the authors via email to obtain category-specific information [19,21]. The extracted data are shown in Supplementary Table 2. We evaluated the risk of bias using the Newcastle-Ottawa Scale (NOS) [23] and the level of evidence using the Oxford Centre for Evidence-Based Medicine (OCEBM) criteria (Supplementary Table 3) [24].

Statistical analysis

We first conducted a meta-analysis for the highest vs. lowest categories of combined UPF intake, as well as analyses of individual UPF subgroups, using the DerSimonian-Laird random-effects models [25,26]. We then conducted linear and nonlinear dose-response meta-analyses for three different UPF units (percentage of g/day, absolute g/day, serving/day).

For linear dose-response meta-analysis, we used the 2-stage generalized least squares trend estimation method [27,28]. Using Greenland and Longnecker’s method [27], we computed the study-specific slopes of linear trends and corresponding 95% CIs from the natural logs of the RRs and CIs extracted across UPF categories. Several approximations have been used to estimate study-specific linear trends. The midpoint of each UPF category was assigned to the corresponding RR. The width of the open-ended extreme categories was assumed to be equivalent to that of the adjacent interval. If the slope of the linear trend was already given in the study, we used the reported slope in the analysis. The study-specific RRs and variances for linear trends were pooled using DerSimonian-Laird random-effects models [26] to estimate the summary RR and 95% CI. Forest plots of the linear dose-response meta-analysis are presented for RRs and 95% CIs associated with each 10% (of g/day), 100-g/day, and 1-serving/day increment in UPF intake. For nonlinear dose-response meta-analysis, we used a 2-stage, random-effects model with restricted cubic splines, modeled with 3 knots, set at the 10th, 50th, and 90th percentiles [29]. The curves derived at each knot were then combined using multivariable random-effects meta-analysis. The P value for nonlinearity was assessed under the null hypothesis that the regression coefficient of the second spline transformation equals zero.

Publication bias was tested using Egger’s test [30] and visually examined using a funnel plot. Sensitivity analyses sequentially omitting each study from the meta-analysis were conducted to check the robustness of results and examine the sources of heterogeneity. We assessed heterogeneity using Cochran’s Q test and the I² statistic [31,32]. Values of I² around 25%, 50%, and 75% were considered low, moderate, and high heterogeneity, respectively [31]. To identify the sources of heterogeneity, subgroup analyses were conducted for potential effect modifiers (region) and methodological characteristics (adjustment for diet quality, total energy intake, and BMI; unit of UPF intake; and timing of dietary assessment). In subgroup analyses by adjustment variables, we restricted the analyses to studies that reported results from both models, with and without adjustment, to enhance comparability between the subgroups (e.g., equal number of studies included and same participant characteristics in both subgroups). For subgroup analysis by diet quality adjustment, we selected the model that was adjusted either by diet quality score (e.g., Lifelines Diet Score, Korean Healthy Eating Index) or nutritional factors (e.g., fiber, saturated fat, sodium, carbohydrate intake). Subgroup analyses were conducted only when there were at least three studies available within each subgroup. All statistical analyses were conducted using STATA version 18.0 (StataCorp., College Station, TX, USA).

RESULTS

After screening 569 publications, 10 publications covering 12 individual prospective studies were identified and included in this dose-response meta-analysis. Of these, five studies originated from Europe [16,17,19-21], four from North America [14,15], and the remaining three from other countries (Korea [11], Brazil [10], Australia [18]). Among them, seven publications received NOS ≥7, indicating high quality, while three publications received NOS 5–6, indicating moderate quality (Supplementary Table 3). The quality of evidence for all studies was rated level 3 by OCEBM (Supplementary Table 3).

Highest vs. lowest categories of UPF intake

A total of 38,308 diabetes cases from 714,199 participants were included in the meta-analysis of the highest vs. lowest categories of combined UPF intake. The summary RR for T2DM risk was 1.48 (95% CI, 1.36 to 1.61), with evidence of moderate to high heterogeneity (I²=73.3%; P<0.001) (Fig. 1). Although there was some heterogeneity in the magnitude of RRs among the studies, all studies consistently showed a positive association. No publication bias was indicated by Egger’s test (P=0.714) (Supplementary Fig. 2).

Among three publications [10,11,15] that reported associations for individual UPF subgroups (for any unit increment), processed meats (summary RR, 1.34; 95% CI, 1.16 to 1.54) and artificially- and sugar-sweetened beverages (1.05; 95% CI, 1.00 to 1.10) were positively associated, whereas ultra-processed cereals and breads (0.98; 95% CI, 0.97 to 0.99), packaged sweet snacks and desserts (0.92; 95% CI, 0.85 to 1.00), and packaged savory snacks (0.92; 95% CI, 0.88 to 0.95) were inversely associated (Table 1).

Among nine studies that reported both models with and without the adjustment for BMI, the positive association was attenuated but remained statistically significant in the BMI-adjusted models (RR, 1.29; 95% CI, 1.18 to 1.42) (Table 2). The summary RRs were similar in the models with and without adjustment for diet quality or total energy intake. In the subgroup analysis by region, higher summary RRs were observed among studies from Europe (summary RR, 1.55; 95% CI, 1.36 to 1.77) and North America (1.55; 95% CI, 1.33 to 1.82) compared with those from other regions (1.29; 95% CI, 1.17 to 1.42; between-subgroup P heterogeneity=0.033). The summary RRs for the highest versus lowest categories of UPF intake were also similar across studies that used different UPF units and between the studies that used baseline versus updated UPF assessments during follow-up.

Percentage of grams per day of UPF intake

Eight studies, including 22,555 diabetes cases from 393,331 participants, reported associations using the percentage of total grams per day of food intake as the unit of UPF intake. Each 10% (of g/day) increment in UPF intake was statistically significantly associated with a 14% higher risk of T2DM (95% CI, 1.11 to 1.17), with evidence of moderate to high heterogeneity (I²=69.2%; P=0.002) (Fig. 2A). The magnitudes of association ranged from 1.03 to 1.25. The associations were similar in both models with and without adjustment for BMI (between-subgroup P heterogeneity=0.203) (Supplementary Fig. 3A) and adjustment for diet quality (between-subgroup P heterogeneity=0.805) (Supplementary Fig. 3B). In subgroup analysis by region, the summary RR was higher in studies from European countries (RR, 1.22; 95% CI, 1.16 to 1.27) compared with that from the United States (1.11; 95% CI, 1.09 to 1.13; between-subgroup P heterogeneity <0.001) (Supplementary Fig. 3C). Egger’s test showed no evidence of publication bias (P=0.183). In the nonlinear dose-response meta-analysis, we found no evidence of nonlinearity (P nonlinearity=0.945) (Fig. 2B).

Absolute grams per day of UPF intake

Four studies, including 14,761 T2DM cases from 300,808 participants, reported associations using absolute grams per day as the unit of UPF intake. Each 100-g/day increment in UPF intake was statistically significantly associated with a 5% higher risk of T2DM (95% CI, 1.03 to 1.06) (Fig. 2C), with evidence of low to moderate heterogeneity (I²=38.5%; P=0.181). No publication bias was detected by Egger’s test (P=0.653). In the nonlinear dose-response meta-analysis, there was evidence of nonlinearity, suggesting a steeper increase in risk approximately at >300 g/day (P nonlinearity=0.001) (Fig. 2D).

Serving per day of UPF intake

Four studies, including 20,690 T2DM cases among 186,467 participants, reported associations using serving per day as the unit of UPF intake. Each one serving/day increase in UPF intake was statistically significantly associated with a 4% higher risk of T2DM (95% CI, 1.03 to 1.05) (Fig. 2E), with evidence of low heterogeneity (I²=21.3%; P=0.283). There was no small study effect, such as publication bias, detected by Egger’s test (P=0.598). Subgroup analysis by adjustment for BMI showed a slightly attenuated association in models with BMI adjustment (RR, 1.02; 95% CI, 1.02 to 1.03 adjusted vs. RR, 1.04; 95% CI, 1.03 to 1.05 unadjusted; between-subgroup P heterogeneity=0.004) (Supplementary Fig. 4). There was no evidence of nonlinearity (P nonlinearity=0.949) (Fig. 2F).

Results were robust in the influence analysis that excluded one study at a time (Supplementary Fig. 5).

DISCUSSION

Our dose-response meta-analysis of 12 prospective cohort studies showed a positive association between UPF intake and T2DM risk. Higher UPF intake (highest vs. lowest categories) was associated with a 48% higher risk of diabetes. The positive association remained statistically significant after additional adjustment for diet quality, total energy intake, or BMI, suggesting that the association is independent of these variables. The summary RRs were also similar between the studies that used baseline vs. updated UPF exposures and among studies that used different UPF units. However, there was statistically significant heterogeneity in the association by region. When the associations were separately estimated for specific UPF units, every 10% (of g/day), 100-g/day, and 1-serving/day increase in UPF intake were associated with 14%, 5%, and 4%, respectively, higher risks of diabetes. Our findings also suggest a possible nonlinear relationship (e.g., threshold effect), showing a steeper risk elevation occurring approximately at >300 g/day of UPF intake.

While the previous meta-analysis focused only on the analysis of single UPF unit (percentage of total g/day) [15], the current study conducted linear and nonlinear dose-response meta-analyses using three different UPF units (percentage of g/day, absolute g/day, and serving/day). Our findings indicated positive linear associations for the percentage of g/day and serving/day. However, evidence of nonlinear relationship was observed with g/day, showing a steeper risk elevation approximately at >300 g/day UPF intake. This finding suggests a potential threshold effect at 300 g/day, which is equivalent to about one can of soda (250 to 350 mL). Furthermore, a noteworthy finding is that, for all three UPF units, we observed no upper limit of risk elevation, indicating that the risk may continue to increase with higher consumption.

In our subgroup analyses by region, we observed a statistically significant regional difference in the association, showing higher summary RRs in Europe and North America than in other regions. The possible explanations for regional differences include discrepancies in UPF composition. Compared with other regions [11,33], European and North American populations have greater consumption of processed meats [34,35], the UPF subgroup that showed the highest summary RR with T2DM risk. The variation in associations among individual UPF subgroups suggest that not all UPFs are equally harmful. Different UPF subgroups have varying nutritional quality and amount of harmful food additives. Processed meats, such as ham and sausage, are high in sodium and saturated fat, which can promote insulin resistance [36] and inflammation [37]. Studies also suggest that the exposure to nitrites and nitrates, the food additives that are mainly observed in processed meats, may increase diabetes risk [38,39]. In contrast, some UPF subgroups, including ultra-processed cereals and breads, packaged sweet snacks and desserts, and packaged savory snacks, were inversely associated in the current meta-analysis. The inverse association with ultra-processed cereals and breads is likely to be driven by the inverse association with ultra-processed whole-grain breads [15]. In a previous meta-analysis of six prospective studies [40], chocolate consumption (highest vs. lowest categories) was associated with a 18% reduced diabetes risk, while the dose-response relationship suggested no reduced risk observed beyond >6 serving/day intake. The inverse association with sweet snacks and desserts may be partially explained by flavanols in chocolates and cocoa products [41,42] and dietary fiber in fruit-based desserts [43]. Flavanols can improve insulin sensitivity [44] by improving the function of pancreatic beta cells and insulin signaling in hepatic cells [45]. Further studies are needed to investigate the biological mechanisms that may explain the link between specific UPF items and disease risk.

In our subgroup analyses by adjustment variables, summary RR was attenuated after adjustment for BMI, suggesting that the association may be at least partially mediated by BMI. As UPFs are typically high in energy density yet less satiating, their consumption often leads to overeating and subsequent weight gain [6,7,46]. Excessive body fat may also promote insulin resistance, leading to an increased risk of T2DM [47,48]. When we compared summary RRs between models with and without adjustment for diet quality or total energy intake, the association did not materially change after the adjustment, suggesting that the association may be independent of these factors. Previous studies have shown that various chemical compounds in UPFs, such as artificial sweeteners and emulsifiers, may disturb the gut microbiota-host relationship and promote insulin resistance without contributing to the total calorie or nutrient content of the food [12,49,50]. It is also possible that, in some populations with small variation in diet quality (e.g., poor diet quality in most individuals), the adjustment for diet quality may produce a little change in the results, limiting our ability to assess the role of diet quality in the association. Therefore, further investigations are needed to confirm the associations in populations with wider ranges of diet quality.

Our study has several limitations. Although we included fully adjusted models in our analysis, we cannot entirely rule out the possibility of unmeasured confounding in individual studies. In addition, our study may be subject to publication bias because we included published studies only. However, Egger’s test indicated no evidence of publication bias. Eight studies reported associations based on baseline UPF intake despite potential changes in intake during follow-up. However, results were similar between the studies that used baseline vs. updated exposures. Lastly, most studies included in our analysis used food frequency questionnaire (FFQ) to assess UPF intake. Because these FFQs were not originally designed to capture UPFs and may lack sufficient information to identify Nova groups (e.g., level of processing), the estimation of UPF intake in these studies may include measurement errors.

Despite these limitations, our study has important strengths. We examined both linear and nonlinear dose-response relationships for three different units of UPF intake, providing more comprehensive estimation of associations compared with the previous meta-analysis that focused on a single unit (percentage of g/day) [15]. Further, while the previous meta-analysis [15] included studies from Western countries only, we included several additional studies that were recently published from non-Western countries. The inclusion of diverse populations enhanced our understanding of this association. Furthermore, we examined the associations of individual UPF subgroups and identified the subgroup with the most adverse effects.

In summary, our dose-response meta-analysis confirmed a positive association between UPF intake and T2DM risk using three different UPF units. The association was also independent of diet quality, total energy intake, and BMI. Given the rapid increase in UPF consumption worldwide, more efforts should focus on reducing consumption of UPFs, particularly processed meats and sugar-sweetened beverages, by increasing accessibility of healthy unprocessed or minimally processed foods and providing support for industry actions to reformulate products. Lastly, more studies are also needed to further investigate the adverse health effects of UPF and its various food additives.

SUPPLEMENTARY MATERIALS

NOTES

CONFLICTS OF INTEREST

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

AUTHOR CONTRIBUTIONS

Conception or design: H.O.

Acquisition, analysis, or interpretation of data: all authors.

Drafting the work or revising: all authors.

Final approval of the manuscript: all authors.

FUNDING

This work was supported by National Research Foundation of Korea grants (RS-2025-00563263; H.O., Y.K.; NRF-2023S1A5 C2A03095169; H.O.) and Korea University grant (K2402691; H.O.). The study funder was not involved in the design of the study; the collection, analysis, and interpretation of data; writing the report; and did not impose any restrictions regarding the publication of the report.

ACKNOWLEDGMENTS

None

Fig. 1.

Forest plot for the association between combined ultra-processed food (UPF) intake (highest vs. lowest categories) and type 2 diabetes mellitus risk. This figure shows results from the meta-analysis comparing the highest and lowest categories of UPF intake using random-effects meta-analyses (Egger’s P=0.714; I²=73.3%; P heterogeneity <0.001). The black diamonds and horizontal lines represent the study-specific relative risk (RR) and their 95% confidence interval (CI) of type 2 diabetes mellitus, and the weights of each of the studies are represented by the size of the gray square. The overall effect estimate and corresponding 95% CI is represented by a hollow diamond. DerSimonian-Laird random-effects models were used for estimating the overall effect. When multiple multivariable models were reported in the study, the estimates from the most adjusted model were used in our meta-analysis. If the study reported multivariable models both with and without adjustment for body mass index (BMI), we included the estimate from the model without BMI in the analysis because BMI can be a potential mediator. When estimates from multiple UPF units were reported, we used the results from the main study model of individual studies (percentage of grams per day from the Korean Genome and Epidemiology Study [KoGES] Ansan-Ansung; serving/day from the Nurses’ Health Study [NHS], Nurses’ Health Study II [NHSII], Health Professionals Follow Up Study [HPFS]). ALSWH, Australian Longitudinal Study on Women’s Health; EPIC, European Prospective Investigation into Cancer and Nutrition; ELSA-Brazil, The Brazilian Longitudinal Study of Adult Health; SUN, Seguimiento Universidad de Navarra; DL, DerSimonian-Laird method.

Fig. 2.

Linear and nonlinear dose-response meta-analyses for the association between combined ultra-processed food (UPF) intake (in percentage of g/day, absolute g/day, and serving/day) and type 2 diabetes mellitus risk. This figure shows the result from (A) linear dose-response relationship for every 10% (of g/day) increase in UPF intake (Egger’s P=0.183; I²=69.2%; P heterogeneity= 0.002), (B) nonlinear dose-response relationship with percentage of g/day as the UPF unit (P nonlinearity=0.945), (C) linear dose-response relationship for every 100-g/day increase in UPF intake (Egger’s P=0.653; I²=38.5%; P heterogeneity=0.181), (D) nonlinear dose-response relationship with absolute g/day as the UPF unit (P nonlinearity=0.001). The minimum value of UPF consumption was 107.1 g/day, (E) linear dose-response relationship for every 1-serving/day increase in UPF intake (Egger’s P=0.598; I²=21.3%; P heterogeneity=0.283), and (F) nonlinear dose-response relationship with serving/d as the UPF unit (P nonlinearity=0.949). The black diamonds and horizontal lines from the forest plots represent the study-specific relative risk (RR) and their 95% confidence interval (CI) of type 2 diabetes mellitus. The weights of each of the studies are represented by the size of the gray square. The overall effect estimates and corresponding 95% CIs are represented by hollow diamonds. P heterogeneity was calculated by Cochran’s Q test. P nonlinearity was calculated using the Wald test for the significance of the second spline coefficient. All statistical tests were two-sided. KoGES, Korean Genome and Epidemiology Study; 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; EPIC, European Prospective Investigation into Cancer and Nutrition; ELSA-Brazil, The Brazilian Longitudinal Study of Adult Health; SUN, Seguimiento Universidad de Navarra.

REFERENCES

• 1. Monteiro CA, Cannon G, Lawrence M, da Costa Louzada ML, Pereira Machado P. Ultra-processed foods, diet quality, and health using the NOVA classification system. Rome: Food and Agriculture Organization of the United Nations: 2019.
• 2. Monteiro CA, Cannon G, Levy RB, Moubarac JC, Louzada ML, Rauber F, et al. Ultra-processed foods: what they are and how to identify them. Public Health Nutr 2019;22:936-41.ArticlePubMedPMC
• 3. Martini D, Godos J, Bonaccio M, Vitaglione P, Grosso G. Ultra-processed foods and nutritional dietary profile: a meta-analysis of nationally representative samples. Nutrients 2021;13:3390.ArticlePubMedPMC
• 4. Moubarac JC, Batal M, Louzada ML, Martinez Steele E, Monteiro CA. Consumption of ultra-processed foods predicts diet quality in Canada. Appetite 2017;108:512-20.ArticlePubMed
• 5. Liu J, Steele EM, Li Y, Karageorgou D, Micha R, Monteiro CA, et al. Consumption of ultra-processed foods and diet quality Among U.S. children and adults. Am J Prev Med 2022;62:252-64.ArticlePubMed
• 6. Beslay M, Srour B, Mejean C, Alles B, Fiolet T, Debras C, et al. Ultra-processed food intake in association with BMI change and risk of overweight and obesity: a prospective analysis of the French NutriNet-Sante cohort. PLoS Med 2020;17:e1003256.ArticlePubMedPMC
• 7. Moradi S, Entezari MH, Mohammadi H, Jayedi A, Lazaridi AV, Kermani MA, et al. Ultra-processed food consumption and adult obesity risk: a systematic review and dose-response meta-analysis. Crit Rev Food Sci Nutr 2023;63:249-60.ArticlePubMed
• 8. Srour B, Fezeu LK, Kesse-Guyot E, Alles B, Mejean C, Andrianasolo RM, et al. Ultra-processed food intake and risk of cardiovascular disease: prospective cohort study (NutriNet-Sante). BMJ 2019;365:l1451.PubMedPMC
• 9. Moradi S, Hojjati Kermani MA, Bagheri R, Mohammadi H, Jayedi A, Lane MM, et al. Ultra-processed food consumption and adult diabetes risk: a systematic review and dose-response meta-analysis. Nutrients 2021;13:4410.ArticlePubMedPMC
• 10. Canhada SL, Vigo A, Levy R, Luft VC, da Fonseca MJ, Giatti L, et al. Association between ultra-processed food consumption and the incidence of type 2 diabetes: the ELSA-Brasil cohort. Diabetol Metab Syndr 2023;15:233.ArticlePubMedPMCPDF
• 11. Cho Y, Ryu S, Kim R, Shin MJ, Oh H. Ultra-processed food intake and risk of type 2 diabetes in Korean adults. J Nutr 2024;154:243-51.ArticlePubMed
• 12. Paula Neto HA, Ausina P, Gomez LS, Leandro JG, Zancan P, Sola-Penna M. Effects of food additives on immune cells as contributors to body weight gain and immune-mediated metabolic dysregulation. Front Immunol 2017;8:1478.PubMedPMC
• 13. Cao Y, Liu H, Qin N, Ren X, Zhu B, Xia X. Impact of food additives on the composition and function of gut microbiota: a review. Trends Food Sci Technol 2020;99:295-310.Article
• 14. Sen A, Brazeau AS, Deschenes S, Ramiro Melgar-Quinonez H, Schmitz N. The role of ultra-processed food consumption and depression on type 2 diabetes incidence: a prospective community study in Quebec, Canada. Public Health Nutr 2023;26:2294-303.ArticlePubMed
• 15. Chen Z, Khandpur N, Desjardins C, Wang L, Monteiro CA, Rossato SL, et al. Ultra-processed food consumption and risk of type 2 diabetes: three large prospective U.S. cohort studies. Diabetes Care 2023;46:1335-44.PubMedPMC
• 16. Duan MJ, Vinke PC, Navis G, Corpeleijn E, Dekker LH. Ultra-processed food and incident type 2 diabetes: studying the underlying consumption patterns to unravel the health effects of this heterogeneous food category in the prospective Lifelines cohort. BMC Med 2022;20:7.ArticlePubMedPMCPDF
• 17. Levy RB, Rauber F, Chang K, Louzada ML, Monteiro CA, Millett C, et al. Ultra-processed food consumption and type 2 diabetes incidence: a prospective cohort study. Clin Nutr 2021;40:3608-14.ArticlePubMed
• 18. Pant A, Gribbin S, Machado P, Hodge A, Wasfy JH, Moran L, et al. Ultra-processed foods and incident cardiovascular disease and hypertension in middle-aged women. Eur J Nutr 2024;63:713-25.ArticlePubMedPDF
• 19. Cordova R, Viallon V, Fontvieille E, Peruchet-Noray L, Jansana A, Wagner KH, et al. Consumption of ultra-processed foods and risk of multimorbidity of cancer and cardiometabolic diseases: a multinational cohort study. Lancet Reg Health Eur 2023;35:100771.ArticlePubMedPMC
• 20. Llavero-Valero M, Escalada-San Martin J, Martinez-Gonzalez MA, Basterra-Gortari FJ, de la Fuente-Arrillaga C, Bes-Rastrollo M. Ultra-processed foods and type-2 diabetes risk in the SUN project: a prospective cohort study. Clin Nutr 2021;40:2817-24.ArticlePubMed
• 21. Srour B, Fezeu LK, Kesse-Guyot E, Alles B, Debras C, Druesne-Pecollo N, et al. Ultraprocessed food consumption and risk of type 2 diabetes among participants of the NutriNet-Sante prospective cohort. JAMA Intern Med 2020;180:283-91.ArticlePubMed
• 22. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. JAMA 2000;283:2008-12.ArticlePubMed
• 23. Wells GA, Shea B, O’Connell D, Peterson J, Welch V, Losos M, et al. The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomised studies in meta-analyses. Available from: https://www.ohri.ca/programs/clinical_epidemiology/oxford.asp (cited 2025 Mar 10).
• 24. Center for Evidence-Based Medicine. The Oxford 2011 levels of evidence. Available from: https://www.cebm.ox.ac.uk/resources/levels-of-evidence/ocebm-levels-of-evidence (cited 2025 Mar 10).
• 25. Dettori JR, Norvell DC, Chapman JR. Fixed-effect vs random-effects models for meta-analysis: 3 points to consider. Global Spine J 2022;12:1624-6.ArticlePubMedPMCPDF
• 26. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986;7:177-88.ArticlePubMed
• 27. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992;135:1301-9.ArticlePubMed
• 28. Orsini N, Bellocco R, Greenland S. Generalized least squares for trend estimation of summarized dose: response data. Stata J 2006;6:40-57.ArticlePDF
• 29. Harrell FE Jr, Lee KL, Pollock BG. Regression models in clinical studies: determining relationships between predictors and response. J Natl Cancer Inst 1988;80:1198-202.ArticlePubMed
• 30. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315:629-34.ArticlePubMedPMC
• 31. Huedo-Medina TB, Sanchez-Meca J, Marin-Martinez F, Botella J. Assessing heterogeneity in meta-analysis: Q statistic or I2 index? Psychol Methods 2006;11:193-206.PubMed
• 32. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002;21:1539-58.ArticlePubMed
• 33. Lee JE, McLerran DF, Rolland B, Chen Y, Grant EJ, Vedanthan R, et al. Meat intake and cause-specific mortality: a pooled analysis of Asian prospective cohort studies. Am J Clin Nutr 2013;98:1032-41.ArticlePubMedPMC
• 34. Zeng L, Ruan M, Liu J, Wilde P, Naumova EN, Mozaffarian D, et al. Trends in processed meat, unprocessed red meat, poultry, and fish consumption in the United States, 1999-2016. J Acad Nutr Diet 2019 119:1085-98. e12.ArticlePubMedPMC
• 35. Rohrmann S, Overvad K, Bueno-de-Mesquita HB, Jakobsen MU, Egeberg R, Tjonneland A, et al. Meat consumption and mortality: results from the European Prospective Investigation into Cancer and Nutrition. BMC Med 2013;11:63.PubMedPMC
• 36. Zelber-Sagi S, Ivancovsky-Wajcman D, Fliss Isakov N, Webb M, Orenstein D, Shibolet O, et al. High red and processed meat consumption is associated with non-alcoholic fatty liver disease and insulin resistance. J Hepatol 2018;68:1239-46.ArticlePubMed
• 37. Chai W, Morimoto Y, Cooney RV, Franke AA, Shvetsov YB, Le Marchand L, et al. Dietary red and processed meat intake and markers of adiposity and inflammation: the multiethnic cohort study. J Am Coll Nutr 2017;36:378-85.ArticlePubMedPMC
• 38. Li T, Lu X, Sun Y, Yang X. Effects of spinach nitrate on insulin resistance, endothelial dysfunction markers and inflammation in mice with high-fat and high-fructose consumption. Food Nutr Res 2016;60:32010.ArticlePubMed
• 39. Helgason T, Ewen SW, Ross IS, Stowers JM. Diabetes produced in mice by smoked/cured mutton. Lancet 1982;2:1017-22.ArticlePubMed
• 40. Yuan S, Li X, Jin Y, Lu J. Chocolate consumption and risk of coronary heart disease, stroke, and diabetes: a meta-analysis of prospective studies. Nutrients 2017;9:688.ArticlePubMedPMC
• 41. Grassi D, Desideri G, Mai F, Martella L, De Feo M, Soddu D, et al. Cocoa, glucose tolerance, and insulin signaling: cardiometabolic protection. J Agric Food Chem 2015;63:9919-26.ArticlePubMed
• 42. Grassi D, Desideri G, Ferri C. Protective effects of dark chocolate on endothelial function and diabetes. Curr Opin Clin Nutr Metab Care 2013;16:662-8.ArticlePubMed
• 43. Tsilas CS, de Souza RJ, Mejia SB, Mirrahimi A, Cozma AI, Jayalath VH, et al. Relation of total sugars, fructose and sucrose with incident type 2 diabetes: a systematic review and meta-analysis of prospective cohort studies. CMAJ 2017;189:E711-20.ArticlePubMedPMC
• 44. Lin X, Zhang I, Li A, Manson JE, Sesso HD, Wang L, et al. Cocoa flavanol intake and biomarkers for cardiometabolic health: a systematic review and meta-analysis of randomized controlled trials. J Nutr 2016;146:2325-33.ArticlePubMedPMC
• 45. Cordero-Herrera I, Martin MA, Bravo L, Goya L, Ramos S. Cocoa flavonoids improve insulin signalling and modulate glucose production via AKT and AMPK in HepG2 cells. Mol Nutr Food Res 2013;57:974-85.PubMed
• 46. Fardet A. Minimally processed foods are more satiating and less hyperglycemic than ultra-processed foods: a preliminary study with 98 ready-to-eat foods. Food Funct 2016;7:2338-46.ArticlePubMed
• 47. Golay A, Ybarra J. Link between obesity and type 2 diabetes. Best Pract Res Clin Endocrinol Metab 2005;19:649-63.ArticlePubMed
• 48. Boden G, Chen X, DeSantis RA, Kendrick Z. Effects of age and body fat on insulin resistance in healthy men. Diabetes Care 1993;16:728-33.PubMed
• 49. Suez J, Korem T, Zeevi D, Zilberman-Schapira G, Thaiss CA, Maza O, et al. Artificial sweeteners induce glucose intolerance by altering the gut microbiota. Nature 2014;514:181-6.ArticlePubMedPDF
• 50. Plaza-Diaz J, Pastor-Villaescusa B, Rueda-Robles A, Abadia-Molina F, Ruiz-Ojeda FJ. Plausible biological interactions of low- and non-calorie sweeteners with the intestinal microbiota: an update of recent studies. Nutrients 2020;12:1153.ArticlePubMedPMC

Figure & Data

References

Citations

Citations to this article as recorded by  

About this article

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 Jun 9. doi: 10.4093/dmj.2024.0706. Epub ahead of print.

Received: Nov 09, 2024; Accepted: Dec 20, 2024

DOI: https://doi.org/10.4093/dmj.2024.0706.

Download citation ↓