Abstract
Aims/hypothesis
Understanding the impact of the overall construct of ultra-processed foods on diabetes risk can inform dietary approaches to diabetes prevention. In this study, we aimed to evaluate the association between ultra-processed food consumption and risk of diabetes in a community-based cohort of middle-aged adults in the USA. We hypothesised that a higher intake of ultra-processed foods is associated with a higher risk of incident diabetes.
Methods
The study included 13,172 participants without diabetes at baseline (1987–1989) in the Atherosclerosis Risk in Communities (ARIC) study. Dietary intake was assessed with a 66-item semiquantitative food frequency questionnaire, and foods were categorised by processing level using the Nova classification system. Ultra-processed food was analysed categorically (quartiles of energy-adjusted intake) and continuously (per one additional serving/day). We used Cox regression to evaluate the association of ultra-processed food intake with risk of diabetes with adjustment for sociodemographic characteristics, total energy intake, health behaviours and clinical factors.
Results
Over a median follow-up of 21 years, there were 4539 cases of incident diabetes. Participants in the highest quartile of ultra-processed food intake (8.4 servings/day on average) had a significantly higher risk of diabetes (HR 1.13; 95% CI 1.03, 1.23) compared with participants in the lowest quartile of intake after adjustment for sociodemographic, lifestyle and clinical factors. Each additional serving of ultra-processed food consumed daily was associated with a 2% higher risk of diabetes (HR 1.02; 95% CI 1.00, 1.04). Highest quartile consumption of certain ultra-processed food groups, including sugar- and artificially sweetened beverages, ultra-processed meats and sugary snacks, was associated with a 29%, 21% and 16% higher risk of diabetes, respectively, compared with the lowest quartile.
Conclusions/interpretation
We found that a higher intake of ultra-processed food was associated with higher risk of incident diabetes, particularly sugar- and artificially sweetened beverages, ultra-processed meats and sugary snacks. Our findings suggest interventions reducing ultra-processed food consumption and specific food groups may be an effective strategy for diabetes prevention.
Keywords: ARIC study, Diabetes prevention, Diet and nutrition, Nova classification, Sugar-sweetened beverages, Ultra-processed food
Introduction
Diabetes is a leading cause of death and disability worldwide, and dietary intake is a key modifiable risk factor for diabetes. Widespread interventions that focus on dietary factors contributing to diabetes risk could have a major population-level impact on future diabetes incidence.
Ultra-processed foods are a group of foods that undergo extensive industrial processing to achieve their final form. They are typically energy-dense, easy to digest and absorb, and often contain high levels of salt, added sugar, refined carbohydrates and unhealthy fats [1]. In addition to these components, ultra-processed foods also contain artificial additives, such as food colorants, emulsifiers and preservatives [2]. Concerns about the potential health impact of ultra-processed foods are growing, with higher consumption linked to a range of adverse outcomes, including weight gain, the metabolic syndrome and cardiovascular diseases, as evidenced by both longitudinal studies and randomised trials [3–5]. Despite this, ultra-processed foods are not addressed in the current Diabetes Standards of Care guidelines [6]. There is a clear need for more evidence to bridge the gap between epidemiological findings and practical applications in clinical practice and public health strategies. Furthermore, ultra-processed foods encompass a broad array of products, making it impractical to recommend complete avoidance. Differentiating specific food groups within the ultra-processed category and understanding their individual links to diabetes can help in crafting more targeted and effective policies and strategies. The evidence on subgroups of ultra-processed food is lacking and necessitates further research.
Most of the previous studies on ultra-processed food and the risk of diabetes have been conducted in European populations, which have different consumption levels and food preferences compared with the US population [7–10]. The studies conducted within the USA have been limited in number and have focused on predominantly White populations with relatively high levels of income and education [11]. Additionally, few studies have investigated the association between individual ultra-processed food subgroups and diabetes risk. Given that ultra-processed foods contribute to over half of the total calories consumed in the USA [12], understanding the impact of ultra-processed foods and subgroups of these foods on diabetes development in a more diverse and generalisable sample of the US population is essential for building more robust evidence to inform future policies.
In light of the current gaps, we investigated the associations of ultra-processed food consumption and specific food groups with incident diabetes risk in a community-based cohort of US adults.
Methods
Study design
The Atherosclerosis Risks in Communities (ARIC) study is a prospective cohort that recruited 15,792 middle-aged adults with representation of men and women, and across racial groups, age, and socioeconomic factors of four geographically diverse US communities (Forsyth County, North Carolina; Jackson, Mississippi; suburbs of Minneapolis, Minnesota; and Washington County, Maryland) at baseline. Participants attended in-person study visits in 1987–1989 (visit 1), 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), 2011–2013 (visit 5), 2016–2017 (visit 6), 2018–2019 (visit 7) and 2020 (visit 8). Participants completed questionnaires about their demographic, socioeconomic, and lifestyle information at the baseline visit (visit 1), and BMI was measured at each visit (https://aric.cscc.unc.edu/aric9/researchers/current_and_archived_visit_documents; accessed 25 June 2024). Dietary intake was assessed at visit 1 and visit 3. All participants provided written consent at each visit, and the study protocol was approved by the institutional review board at each study centre [13].
We excluded participants who had more than ten items missing from the food frequency questionnaire (n=364) and those with implausible energy intake (defined as <2510 or >18,828 kJ/day for men and <2092 or >14,644 kJ/day for women) or missing energy intake information (n=13). Subsequently, we excluded participants with missing covariates (n=321); participants who were neither White nor Black (i.e. people who self-reported as Asian, American Indian or Alaskan Indian; n=45), and Black participants from the Minneapolis (n=22) and Washington County (n=31) study centres, both due to limited sample sizes. Additionally, we excluded individuals with prevalent diabetes (defined as having a fasting glucose concentration of 7 mmol/l or higher, a non-fasting glucose concentration of 11.1 mmol/l or higher, a self-reported physician’s diagnosis of diabetes, or the use of diabetes medication in the 2 weeks preceding the baseline visit) (n=1752), or missing incident diabetes information (n=72). The final sample size included 13,172 Black and White participants without diabetes at baseline (electronic supplementary material [ESM] Fig. 1).
Dietary intake assessment and ultra-processed food definition
Participants’ usual eating habits were assessed at visit 1 and visit 3 using a 66-item semiquantitative food frequency questionnaire (FFQ). This FFQ was a modified version of the Harvard Willett FFQ [14]. The questionnaire was completed during the interview portion of the visit and was administered by study staff. Participants were instructed to estimate their consumption frequency for foods in specific portions during the year preceding the study visit. Study staff also provided various size of cups and glasses to help participants visualise portion size and more accurately estimate the quantity of food consumed. After obtaining the consumption frequency of foods, nutrient intake was calculated by linkage to nutritional content data from the US Department of Agriculture [15].
All 66 food items were classified into one of four processing categories based on the Nova classification: (1) minimally processed or unprocessed foods, obtained directly from the plant or animal source with no alteration of the nutritional properties of the food; (2) processed culinary ingredients, extracted from natural sources and used for culinary purposes; (3) processed foods, which contain minimally processed or unprocessed foods and processed culinary ingredients, and undergo processes including drying, canning or the addition of salt, sugar or oil; and (4) ultra-processed foods, which have undergone extensive processing and often contain ingredients not typically used in culinary preparations [16]. Detailed descriptions of each category and the classification of food items in the ARIC study were previously described [5]. We grouped ultra-processed foods into nine subgroups according to their categorisation on the FFQ: sugar- and artificially sweetened beverages, ultra-processed meats, sugary snacks, fried foods, margarine, cereals, hard liquor (spirits), baked goods and ice cream [5].
After summing intakes of foods from each processing category in servings/day, we adjusted for total energy intake using the residual method [17]. Residuals of ultra-processed food were calculated as the difference between observed and predicted values in a regression model with ultra-processed food intake as the dependent variable and total energy intake as the independent variable. Quartiles were generated by ranking energy-adjusted intake of ultra-processed foods.
We modelled ultra-processed food as a time-varying exposure, to incorporate repeated dietary assessment at visit 1 and visit 3. Between visit 1 and visit 3, ultra-processed food consumption was represented by intake at visit 1. For participants who completed the visit 3 FFQ and did not develop diabetes before visit 3, ultra-processed food consumption after visit 3 was calculated as the mean intakes of visit 1 and visit 3. There was minimal change in ultra-processed food consumption from visit 1 to visit 3 (mean change: 0.08 servings/day).
Incident diabetes ascertainment
We followed participants for incident diabetes from baseline (visit 1) through to 31 December 2020. Incident diabetes was defined as either having a fasting glucose ≥7 mmol/l, having a non-fasting glucose ≥11.1 mmol/l, or self-reported physician diagnosis of diabetes or use of glucose-lowering medication. Blood glucose was measured at each study visit. The assessment of participants’ previous history of diabetes diagnosis or medication use was conducted by study staff during the second (1990–1992), third (1993–1995), and fourth (1996–1998) study visits. Following the fourth visit, this information was gathered based on the responses to follow-up phone calls with either participants or their proxies (annually prior to 2012 and twice yearly thereafter) [18].
We conducted a sensitivity analysis limiting incident diabetes to treated case only, i.e. only those cases identified by the self-reported use of glucose-lowering medications at any visit or telephone call during follow-up.
Assessment of covariates
Information on participants’ baseline characteristics was obtained through interviews conducted by the field centre staff. This information included participants’ age, sex, race, education level, smoking status, physical activity score and medical history. Sex was self-reported by participants. Physical activity during leisure time was assessed using the Baecke questionnaire [19]. BMI (kg/m2) was calculated using on-site weight and height measurements. Seated blood pressure measurements were taken three times during the visit, and the latter two readings were averaged in our definition of hypertension status. Prevalent hypertension was defined as having a systolic blood pressure of 140 mmHg or higher, a diastolic blood pressure of 90 mmHg or higher, or taking any antihypertensive medication in the 2 weeks prior to the baseline visit. eGFR was calculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation using age, sex and serum creatinine (without race) [20].
Statistical analysis
We summarised baseline characteristics and nutrient intakes according to quartiles of energy-adjusted ultra-processed food consumption. To assess differences across quartiles, we used χ2 tests for categorical variables and analysis of variance for continuous variables.
We calculated HR and 95% CI for the association between ultra-processed food consumption and the risk of incident diabetes using Cox proportional hazards regression models, with time from study entry to the first occurrence of diabetes or censoring (death, administrative censoring at 31 December 2020) as the time metric. In Model 1, we controlled for age, sex, race–centre (modelled as a combined variable in order to account for the unevenly distributed race groups across study centres), and total energy intake. In Model 2, which we considered as our main model, we additionally controlled for health behaviours, including smoking status and physical activity score; education level, as a proxy for socioeconomic status; and clinical factors, including hypertension status and kidney function (defined by eGFR levels modelled as two linear spline terms with one knot at 90 mL/min per 1.73 m2). Model 3 included Model 2 covariates plus baseline BMI, as a potential mediator of the association between ultra-processed food and diabetes. We also conducted a formal mediation analysis to examine BMI as a mediator. We modelled ultra-processed food intake both categorically (quartiles of energy-adjusted ultra-processed food intake) and continuously (per one additional serving/day). For the categorical analysis, we tested for a linear trend across quartiles using the median energy-adjusted intake within each quartile and reported p values for trend.
To visually examine the risk of incident diabetes across the range of intakes, we modelled both ultra-processed food and minimally processed or unprocessed food intake using restricted cubic splines with four knots, positioned at the 5th, 35th, 65th and 95th percentiles of daily intake, and the 25th percentile set as the reference [21].
We assessed associations between nine ultra-processed food subgroups and the risk of diabetes: sugar- and artificially sweetened beverages, margarine, baked goods, ultra-processed meats, cereals, fried foods, sugary snacks, hard liquor (spirits) and ice cream. We divided participants’ energy-adjusted intakes of each food group into quartiles. We calculated HRs for each food group while controlling for all the other food groups consumption, plus covariates from Model 2.
Sensitivity analyses were carried out by: (1) defining incident diabetes as present in those who reported current use of glucose-lowering medications only (i.e. treated diabetes); (2) adjusting for time-varying BMI instead of baseline BMI, in addition to covariates included in Model 2. Participants’ BMI was updated at each visit when available, and missing values were imputed using the most recent previous measurements; (3) excluding incident diabetes cases that occurred within the first 5 years of follow-up to avoid potential reverse causation; (4) adjusting for diet quality [Alternative Healthy Eating Index 2010 (AHEI-2010)] [22]; (5) using time-updated covariates. All data analyses were performed using Stata version 17.0 (StataCorp, USA).
Results
Baseline characteristics
The mean age was 54 years, and the mean BMI was 27 kg/m2 (Table 1). Participants with the highest consumption of ultra-processed food were more likely to be White, to have at least a high school education, to be a former smoker, and to be obese. Age, sex, physical activity, kidney function and prevalence of hypertension were similar across quartiles of ultra-processed food.
Table 1.
Demographic, socioeconomic, behavioural, clinical and nutritional characteristics according to quartiles of ultra-processed food intake
| Quartile 1 | Quartile 2 | Quartile 3 | Quartile 4 | |
|---|---|---|---|---|
| (N=3293) | (N=3293) | (N=3293) | (N=3293) | |
| Demographic, socioeconomic, behavioural and clinical characteristics | ||||
| Age, years | 54.2 ± 5.7 | 54.0 ± 5.8 | 53.9 ± 5.7 | 53.8 ± 5.7 |
| Female sex | 1773 (53.8) | 1927 (58.5) | 1870 (56.8) | 1729 (52.5) |
| Self-reported Black race | 1003 (30.5) | 934 (28.4) | 769 (23.4) | 393 (11.9) |
| BMI, kg/m2 | 27.0 ± 5.1 | 27.2 ± 5.1 | 27.3 ± 5.1 | 27.5 ± 5.1 |
| BMI categories | ||||
| <25 kg/m2 | 1215 (36.9) | 1220 (37.0) | 1184 (36.0) | 1108 (33.6) |
| ≥25.0–<30 kg/m2 | 1321 (40.1) | 1294 (39.3) | 1314 (39.9) | 1339 (40.7) |
| ≥30 kg/m2 | 757 (23.0) | 779 (23.7) | 795 (24.1) | 846 (25.7) |
| Physical activity scorea | 2.5 ± 0.8 | 2.4 ± 0.8 | 2.4 ± 0.8 | 2.5 ± 0.8 |
| Education level | ||||
| Less than high school | 807 (24.5) | 728 (22.1) | 681 (20.7) | 608 (18.5) |
| High school | 964 (29.3) | 1063 (32.3) | 1123 (34.1) | 1162 (35.3) |
| Higher than high school | 1522 (46.2) | 1502 (45.6) | 1489 (45.2) | 1523 (46.2) |
| Smoking status | ||||
| Current smoker | 960 (29.2) | 844 (25.6) | 825 (25.1) | 837 (25.4) |
| Former smoker | 992 (30.1) | 1059 (32.2) | 1033 (31.4) | 1182 (35.9) |
| Never smoker | 1341 (40.7) | 1390 (42.2) | 1435 (43.6) | 1274 (38.7) |
| eGFR, mL/min per 1.73 m2 | 101.8 ± 12.5 | 102.0 ± 12.7 | 102.1 ± 12.1 | 102.5 ± 11.7 |
| Hypertension statusb | 872 (26.5) | 899 (27.3) | 873 (26.5) | 792 (24.1) |
| Nutritional characteristics | ||||
| Energy-adjusted ultra-processed foods intake, servings/day | 3.6 ± 1.0 | 5.2 ± 0.3 | 6.2 ± 0.3 | 8.4 ± 1.6 |
| Energy-adjusted minimally processed or unprocessed foods intake, servings/day | 10.1 ± 3.3 | 8.8 ± 2.7 | 8.3 ± 2.7 | 7.6 ± 2.9 |
| Alternative Healthy Eating Index 2010 | 52.2 ± 12.5 | 46.4 ± 11.4 | 43.8 ± 11.1 | 44.0 ± 11.0 |
| Total energy intake, kJ/day | 7354.2 ± 2551.4 | 6179.2 ± 2247.9 | 6350.6 ± 2347.2 | 7209.9 ± 2671.1 |
| Food groups intake, servings/day | ||||
| Whole fruits | 1.9 ± 1.7 | 1.4 ± 1.1 | 1.3 ± 1.0 | 1.2 ± 1.0 |
| Vegetables | 2.2 ± 1.4 | 1.7 ± 1.1 | 1.5 ± 1.0 | 1.5 ± 1.0 |
| Dairy products | 3.0 ± 2.1 | 2.6 ± 1.8 | 2.8 ± 1.8 | 3.7 ± 2.2 |
| Red meats | 1.3 ± 1.0 | 1.2 ± 0.8 | 1.2 ± 0.8 | 1.4 ± 1.0 |
| Sugar- and artificially sweetened beverages | 1.0 ± 0.9 | 1.2 ± 0.9 | 1.6 ± 1.1 | 2.5 ± 1.8 |
| Alcohol intake, g/day | 7.2 ± 15.4 | 5.3 ± 11.3 | 5.6 ± 11.4 | 7.2 ± 16.2 |
| Macronutrient intake, % of energy | ||||
| Protein | 19.3 ± 4.3 | 18.4 ± 3.9 | 17.2 ± 3.7 | 16.1 ± 3.8 |
| Carbohydrate | 48.6 ± 9.4 | 49.0 ± 8.9 | 49.4 ± 9.2 | 48.9 ± 9.8 |
| Total fat | 31.6 ± 6.9 | 32.3 ± 6.3 | 33.0 ± 6.5 | 34.3 ± 6.9 |
| Saturated fat | 11.7 ± 3.2 | 11.8 ± 2.9 | 12.1 ± 2.9 | 12.4 ± 3.0 |
| Fibre, g/1000 kJ | 2.9 ± 1.1 | 2.7 ± 1.0 | 2.5 ± 0.9 | 2.3 ± 0.8 |
Baseline characteristics are reported as mean ± SD for continuous variables and number (%) for categorical variables
Physical activity score was calculated for sport-related exercise during leisure time and has a range from 1 to 5
Hypertension status was defined as systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥90 mmHg, or use of antihypertensive medication during the past 2 weeks before visit
On average, participants in quartile 4 consumed 8.4 servings/day of ultra-processed foods vs 3.6 servings/day in quartile 1, and 7.6 servings/day of minimally processed or unprocessed foods vs 10.1 servings/day in quartile 1 (Table 1). Those who consumed more ultra-processed foods ate fewer whole fruits and vegetables and drank more sugar- and artificially sweetened beverages. They also consumed less protein and fibre, and more total fat and saturated fat.
The top three food groups contributing to ultra-processed food consumption were sugar- and artificially sweetened beverages, baked goods, and margarine (ESM Fig. 2).
Ultra-processed food and incident diabetes risk
Over a median follow-up of 21 years, there were 4539 incident diabetes cases. Diabetes risk increased across quartiles of ultra-processed food intake (Table 2). Participants in the highest quartile of ultra-processed food intake had a significantly higher risk of incident diabetes compared with those in the first quartile (Model 1: HR 1.17; 95% CI 1.07, 1.27). The association was attenuated but remained significant after further adjustment for education, smoking status, physical activity, hypertension status, and kidney function (Model 2: HR 1.13; 95% CI 1.03, 1.23), The observed association remained statistically significant in Model 2 after correcting for multiple comparisons (p<0.05/3=0.017).
Table 2.
Risk of diabetes associated with ultra-processed and minimally processed or unprocessed food consumption
| Quartile 1 (N=3293) | Quartile 2 (N=3293) | Quartile 3 (N=3293) | Quartile 4 (N=3293) | p value for trend | Per 1 additional serving of ultra-processed food per day | p value | |
|---|---|---|---|---|---|---|---|
| Ultra-processed foods | |||||||
| Events observed, n (%) | 1090 (33.1) | 1150 (34.9) | 1137 (34.5) | 1162 (35.3) | |||
| Incidence rate per | 16.9 (15.9, 17.9) | 17.4 (16.5, 18.5) | 17.1 (16.1, 18.1) | 17.4 (16.4, 18.4) | |||
| 1000 person-years (95% CI) | |||||||
| Model 1a | 1 [reference] | 1.12 (1.03, 1.22) | 1.14 (1.04, 1.24) | 1.17 (1.07, 1.27) | <0.001 | 1.03 (1.02, 1.04) | <0.001 |
| Model 2b | 1 [reference] | 1.09 (1.01, 1.19) | 1.10 (1.01, 1.20) | 1.13 (1.03, 1.23) | 0.008 | 1.02 (1.00, 1.04) | 0.002 |
| Model 3c | 1 [reference] | 1.08 (0.99, 1.18) | 1.08 (0.99, 1.17) | 1.06 (0.97, 1.16) | 0.2 | 1.01 (1.00, 1.03) | 0.1 |
| Minimally processed or unprocessed foods | |||||||
| Events observed, n (%) | 1223 (37.1) | 1167 (35.4) | 1129 (34.3) | 1020 (31.0) | |||
| Incidence rate per 1000 person-years (95% CI) | 19.2 (18.1, 20.3) | 17.8 (16.8, 18.8) | 16.9 (16.0, 17.9) | 15.0 (14.1, 16.0) | |||
| Model 1a | 1 [reference] | 0.96 (0.88, 1.03) | 0.89 (0.82, 0.97) | 0.87 (0.80, 0.95) | <0.001 | 0.98 (0.97, 0.99) | <0.001 |
| Model 2b | 1 [reference] | 0.99 (0.91, 1.07) | 0.93 (0.86, 1.02) | 0.91 (0.83, 0.99) | 0.006 | 0.98 (0.97, 0.99) | 0.002 |
| Model 3c | 1 [reference] | 0.99 (0.91, 1.07) | 0.92 (0.85, 1.00) | 0.90 (0.82, 0.98) | 0.003 | 0.98 (0.97, 0.99) | 0.001 |
Model 1 was adjusted for age, sex, race–centre and total energy intake
Model 2 was adjusted for variables in Model 1 plus smoking status, physical activity score, education level, hypertension status, and kidney function (estimated glomerular filtration rate, two linear spline terms with one knot at 90 mL/min per 1.73 m2)
Model 3 was adjusted for variables in Model 2 plus baseline BMI
When we additionally adjusted for baseline BMI, the association between ultra-processed food and the risk of incident diabetes was attenuated and no longer statistically significant (Model 3: HR 1.06; 95% CI 0.97, 1.16). BMI mediated 40% of the association between ultra-processed food and diabetes (average causal mediation effect=0.0027, p<0.001).
The association between ultra-processed food intake (servings per day) and the risk of incident diabetes was roughly linear (Fig. 1a), with a 2% higher risk for each additional serving/day of ultra-processed foods consumed (Model 2: HR 1.02; 95% CI 1.00, 1.04, Table 2).
Fig. 1.
Association between ultra-processed (a) and minimally processed or unprocessed (b) food intake and risk of incident diabetes represented by restricted cubic splines. The model was adjusted for age, sex, race–centre, total energy intake, smoking status, physical activity score, education level, hypertension status and kidney function (two linear spline terms with one knot at 90 ml/min per 1.73 m2). The black solid line represented the adjusted HR of incident diabetes. The black dashed lines represent the 95% CI. The grey histogram represents the frequency of ultra-processed food consumption (a) and minimally processed or unprocessed food consumption (b) in the study population. For ultra-processed foods, the reference level was set at the 25th percentile (3.9 servings per day), and 4 knots were set at the 5th, 35th, 65th and 95th percentiles (2.2, 4.5, 6.4 and 10.9 servings per day, respectively). For minimally processed or unprocessed foods, the reference level was set at the 25th percentile (6.3 servings per day), and 4 knots were set at the 5th, 35th, 65th and 95th percentiles (4.0, 7.2, 9.7 and 14.9 servings per day, respectively)
Minimally processed or unprocessed food and incident diabetes risk
The incidence of diabetes was lower for those with higher intake of minimally processed or unprocessed food (Table 2). Participants with the highest quartile of minimally processed or unprocessed food consumption had a lower risk of incident diabetes compared to those with the lowest consumption (Model 1: HR 0.87; 95% CI 0.80, 0.95). The association was attenuated but remained significant in the main model after controlling for health behaviours, socioeconomics and clinical factors (Model 2: HR 0.91; 95% CI 0.83, 0.99). The association remained significant after accounting for baseline BMI (Model 3: HR 0.90; 95% CI 0.82, 0.98).
There was a dose–response relationship between higher consumption of minimally processed or unprocessed food and lower risk of diabetes (Fig. 1b), with a 2% lower risk of diabetes per each serving of minimally processed or unprocessed food (Model 2: HR 0.98; 95% CI 0.97, 0.99, Table 2).
Associations for minimally processed or unprocessed food and diabetes were significant after accounting for multiple comparisons for the three models (p<0.05/3=0.017).
Ultra-processed food groups and risk of incident diabetes
Higher intakes of sugar- and artificially sweetened beverages, ultra-processed meats and sugary snacks were associated with a higher risk of diabetes (HR 1.29, 1.21, and 1.16, respectively), for quartile 4 vs quartile 1 (Fig. 2, ESM Table 1). Conversely, baked goods (HR 0.88; 95% CI 0.77, 0.99) and ice cream (HR 0.88; 95% CI 0.79, 0.97) intake were inversely associated with diabetes risk.
Fig. 2.
Risk of incident diabetes associated with consumption of specific ultra-processed food items or groups. HR for incident diabetes were calculated for highest consumption (quartile 4) vs lowest consumption (quartile 1) of the specific ultra-processed food items or groups. Model included all food groups simultaneously while adjusting for age, sex, race–centre, total energy intake, smoking status, physical activity score, education level, hypertension status and kidney function (two linear spline terms with 1 knot at 90 ml/min per 1.73 m2). aSugar- and artificially sweetened beverages included orange or grapefruit juice, low energy soft drinks (e.g. Diet Coke, Diet Pepsi, 7 Up Zero Sugar) and regular soft drinks (e.g. Coca-Cola, Pepsi, 7 Up, ginger ale), fruit-flavoured punch or non-carbonated beverages (e.g. lemonade, Kool-Aid, Hawaiian Punch; not diet). bUltra-processed meats included hamburgers, hot dogs, processed meats (e.g. sausage, salami, bologna); beef, pork or lamb in dishes. cSugary snacks included chocolate bars or pieces (e.g. Hershey’s, Plain M&M’s, Snickers, Reese’s), candy without chocolate. dFried foods included potato chips or corn chips, French fried potatoes, food fried away from home. eBaked goods included ready-made pie, doughnuts, biscuits or cornbread; Danish pastry, sweet roll, coffee cake and croissant; cookies, cake or brownie. Hard liquor is also termed ‘spirits’ (e.g. vodka, rum, gin, whisky, tequila)
Sensitivity analysis
The magnitude of our results was similar when we limited the outcome to cases of medication-treated diabetes, but the estimates were less precise and some were no longer significant (ESM Table 2). In other sensitivity analyses, the results were largely consistent with the following exceptions: ultra-processed food intake was no longer significant after adjusting for time-varying BMI, and minimally processed or unprocessed foods became non-significant after adjusting for AHEI-2010 (ESM Table 3).
Discussion
In this large prospective cohort of middle-aged men and women, higher intake of ultra-processed food was associated with an elevated risk of diabetes. Higher intake of minimally processed or unprocessed food was associated with a lower diabetes risk. Our analysis reveals a dose–response relationship between increased servings of ultra-processed foods and higher diabetes risk, with no evidence of a ‘safe’ consumption threshold. Among specific ultra-processed food groups, sugar- and artificially sweetened beverages, sugary snacks and ultra-processed meats were the most strongly associated with higher risk of diabetes.
Our findings align with previous research. The USA-based Nurses’ Health Study, Nurses’ Health Study II (NHS & NHSII), and the Health Professionals Follow-Up Study (HPFS) reported HRs ranging from 1.19 to 1.46 when comparing the highest to the lowest quintiles of ultra-processed food consumption [11]. This previous study conducted in NHS, NHSII and HPFS also highlighted higher diabetes risk associated with sugar- and artificially sweetened beverages and animal-based products, as well as a lower diabetes risk associated with dairy-based desserts, including ice cream. Our findings were also consistent with results from several European-based population studies, which reported a 15–55% increased risk of diabetes associated with greater ultra-processed food consumption (either comparing the highest vs lowest consumption groups or by a 10% increment of ultra-processed foods in the diet) [7–9].
The elevated risk of diabetes from higher ultra-processed food consumption may be attributed to several underlying pathways. We found that adjusting for BMI diminished the association between ultra-processed food consumption and diabetes risk. Weight gain and obesity are critically important risk factors for diabetes, and ultra-processed food consumption has been consistently linked with both weight gain and obesity [23]. The energy density, texture, hyper-palatability and ease of preparation of ultra-processed food can contribute to overeating and subsequent weight gain [24]. In a randomised feeding study, participants consumed more calories from an ultra-processed diet compared with an unprocessed diet and experienced a significant increase in body weight over a 2-week period [3]. Our results suggest that body weight may play an important mediating role in the association between ultra-processed food and risk of diabetes.
Ultra-processed foods are low in nutritional value and are often characterised by high levels of added sugar and saturated fat, and low fibre content [25]. Added sugar is associated with insulin resistance, inflammation and obesity [26]. Previous evidence has recommended replacing saturated fat with unsaturated fat for type 2 diabetes prevention [27]. Dietary intake of saturated fat can increase intrahepatic triacylglycerol content, leading to nonalcoholic fatty liver disease which is associated with impaired glucose metabolism [28]. On the other hand, a higher intake of fibre has been shown to be associated with a lower risk of type 2 diabetes [29]. In the current study, we found that higher consumption of ultra-processed food was associated with reduced intake of fruits, vegetables, and fibre, indicating that healthier foods were replaced by ultra-processed options, leading to diabetes risk.
Some other potential pathways that may contribute to the observed associations include food additives (e.g. preservatives, artificial sweeteners, colorants) used to enhance flavour and/or prolong shelf life, and the unintended introduction of chemical compounds during the food processing and packaging processes (i.e. neo-formed compounds). Propionate, which is used in breakfast cereals and sausage casings as a preservative, may contribute to insulin resistance by stimulating glycogenolysis and increasing plasma glucagon and fatty acid-binding protein 4 (FABP4) levels [30]. Artificial sweeteners like aspartame and sucralose, common in ultra-processed foods and beverages as sugar substitutes, have been linked to higher type 2 diabetes risk in experimental and epidemiological studies [31]. Acrylamides, heterocyclic amines and 5-hydroxymethylfurfural, formed during processing, along with packaging contaminants like phthalates, bisphenol A and perfluoroalkyl substances, are linked to endocrine disruption, inflammation, insulin resistance and increased diabetes risk [32–34].
Our study also revealed significant associations between various ultra-processed food groups and the risk of incident diabetes. Notably, higher consumption of sugar- and artificially sweetened beverages was linked to a 29% higher risk of diabetes, echoing the abundant evidence in the literature regarding the detrimental impact of these beverages on diabetes risk [35]. The observed association with sugar- and artificially sweetened beverages could also be due to their high glycaemic load [36]. Our findings also indicated that those with the highest consumption of ultra-processed meats had a 21% higher risk of diabetes. Ultra-processed meats are often rich in saturated fats, sodium, nitrites and nitrates, which could impact metabolic health and the risk of diabetes [37, 38]. A meta-analysis graded the evidence as high for the associations between sugar-sweetened beverages, red meat and processed meat and the risk of diabetes [39]. These results strengthen the case for more specific recommendations within diabetes prevention guidelines, suggesting a reduction in the consumption of sugar- and artificially sweetened beverages and foods and ultra-processed meats as a strategy to prevent or delay the onset of diabetes.
Surprisingly, a higher intake of baked goods and ice cream was associated with a lower risk of incident diabetes. For baked goods, it is possible that people were consuming healthier variants enriched with fibre and whole grains, which could mitigate the risk of diabetes. Regarding ice cream, a similar inverse association has been reported in previous cohort studies [11, 40, 41]. The high dairy fat content in ice cream might contribute to this association. Another possible explanation for both observations is reverse causation, where individuals already at a higher risk of diabetes (e.g. those with prediabetes) might consciously avoid consuming baked goods and ice cream, thereby skewing the true association between these foods and diabetes.
Our study has several strengths. First, the prospective design and extensive follow-up of the ARIC study enabled us to assess the temporal relationship between ultra-processed food intake and future risk of diabetes. Second, our study included both Black and White participants across four US communities, which enhances the external validity of our findings. Third, we incorporated time-updated estimates of ultra-processed food intake by using repeated dietary assessments to provide a more accurate estimate of participants’ dietary intake over time. Lastly, the ARIC study rigorously ascertained incident diabetes cases. By examining two different definitions of diabetes, we confirmed the robustness of the associations observed in our study.
Nevertheless, several limitations should be acknowledged. First, the FFQ used was not specifically designed to capture ultra-processed food consumption and was administered in the late 1980s and mid-1990s. The ultra-processed food items captured may not fully represent the current food supply due to changes over time in dietary habits caused by societal shifts (e.g. margarine consumption) and individual changes (e.g. ageing, employment, family composition, comorbidities). Further research is necessary using contemporary dietary data. Second, there is the possibility of misclassification due to the lack of detailed information provided for some food items. Due to the lack of a gold standard, we did not conduct a validation study on classifying ultra-processed food. To enhance rigour, two individuals independently classified food items and a third resolved discrepancies. Third, although the ARIC study provided a comprehensive and rigorous assessment of participants’ characteristics, and we adjusted for a wide range of demographic, socioeconomic, clinical and lifestyle variables, residual confounding remains a concern. Ultra-processed food consumption and diabetes risk can be affected by stress, occupation and income. Future studies with these measures are needed. Fourth, the observational nature of the study raises the possibility of reverse causation, especially considering the long-term follow-up. To address this, we performed sensitivity analyses excluding early incident cases (i.e. those cases that were identified within the first 5 years of follow-up). Fifth, we are unable to distinguish between types of diabetes (gestational, type 1 or type 2) in the ARIC study. However, since we excluded prevalent diabetes at baseline, and the participants were 45–65 years old at enrolment, it is likely that most cases were type 2 diabetes.
In conclusion, our study showed that greater intake of ultra-processed food was associated with a higher risk of incident diabetes in middle-aged Black and White men and women. Our findings add to the evidence linking ultra-processed food intake with diabetes risk and suggest potential heterogeneity in the associations between different types of ultra-processed foods and diabetes risk. Dietary guidance should recommend avoiding the consumption of ultra-processed food and specifically sugar-sweetened and artificially sweetened foods and beverages and ultra-processed meat and emphasise the benefits of increasing intake of minimally processed or unprocessed foods in order to mitigate diabetes risk.
Supplementary Material
Research in context.
What is already known about this subject?
Ultra-processed foods consist of non-culinary substances, have undergone extensive processing and are typically low in nutritional quality
Ultra-processed foods comprise a dominant portion of our diet and have been linked with several health outcomes, including weight gain and cardiometabolic diseases
What is the key question?
What is the extent of the association between ultra-processed food intake and the risk of incident diabetes in a diverse population, and what are the associations between specific food groups and diabetes risk?
What are the new findings?
High ultra-processed food consumption was associated with a 13% higher diabetes risk, and each daily serving of ultra-processed food was associated with a 2% higher risk of diabetes in a racially diverse US study population
There appears to be a dose-response relationship between servings of ultra-processed food and the risk of incident diabetes, with no clear threshold for ‘safe’ consumption
Sugar- and artificially-sweetened beverages, ultra-processed meats and sugary snacks are the top three food groups with the strongest associations with incident diabetes risk
How might this impact on clinical practice in the foreseeable future?
Our findings suggest that minimising the consumption of ultra-processed foods may be a key step in preventing diabetes. Future guidelines should consider incorporating the evidence on ultra-processed food in order to reduce consumption of such foods
Acknowledgements
The authors thank the staff and participants of the ARIC study for their important contributions.
Funding
The Atherosclerosis Risk in Communities study has been funded in whole or in part with federal funds from the National Heart, Lung, and Blood Institute, National Institutes of Health, Department of Health and Human Services (75N92022D00001, 75N92022D00002, 75N92022D00003, 75N92022D00004, 75N92022D00005). CMR was supported by a grant from the National Heart, Lung, and Blood Institute (R01 HL153178). ES was supported by NIH/NHLBI grant K24 HL152440. MF was supported by NIH/NIDDK grant K01-DK138273. The study sponsor/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. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Abbreviations
- ARIC
Atherosclerosis Risk in Communities
- FFQ
Food frequency questionnaire
- HPFS
Health Professionals Follow-Up Study
Footnotes
Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s00125-024-06221-5.
Author’s relationships and activities ES is a member of the editorial board of Diabetologia but was not involved in any of the decisions regarding review of the manuscript or its acceptance for publication. All other authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.
Data availability
Data described in the manuscript are available upon request pending application and approval from the National Heart, Lung, and Blood Institute (NHLBI) Biologic Specimen and Data Repository Information Coordinating Center (BioLINCC) (https://biolincc.nhlbi.nih.gov/home/) or the Atherosclerosis Risk in Communities study.
References
- 1.Martini D, Godos J, Bonaccio M, Vitaglione P, Grosso G (2021) Ultra-processed foods and nutritional dietary profile: a meta-analysis of nationally representative samples. Nutrients 13(10):3390. 10.3390/nu13103390 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Monteiro CA, Cannon G, Levy RB et al. (2019) Ultra-processed foods: what they are and how to identify them. Public Health Nutr 22(5):936–941. 10.1017/S1368980018003762 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Hall KD, Ayuketah A, Brychta R et al. (2019) Ultra-processed diets cause excess calorie intake and weight gain: an inpatient randomized controlled trial of ad libitum food intake. Cell Metab 30(1):67–77.e3. 10.1016/j.cmet.2019.05.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Canhada SL, Vigo Á, Luft VC et al. (2023) Ultra-processed food consumption and increased risk of metabolic syndrome in adults: The ELSA-Brasil. Diabetes Care 46(2):369–376. 10.2337/dc22-1505 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Du S, Kim H, Rebholz CM (2021) Higher Ultra-processed food consumption is associated with increased risk of incident coronary artery disease in the atherosclerosis risk in communities study. J Nutr 151(12):3746–3754. 10.1093/jn/nxab285 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.American Diabetes Association Professional Practice Committee (2024) 3. Prevention or delay of diabetes and associated comorbidities: Standards of care in diabetes-2024. Diabetes Care 47(Suppl 1):S43–S51. 10.2337/dc24-S003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Llavero-Valero M, Escalada-San Martín J, Martínez-González MA, Basterra-Gortari FJ, de la Fuente-Arrillaga C, Bes-Rastrollo M (2021) Ultra-processed foods and type-2 diabetes risk in the SUN project: a prospective cohort study. Clin Nutr 40(5):2817–2824. 10.1016/j.clnu.2021.03.039 [DOI] [PubMed] [Google Scholar]
- 8.Srour B, Fezeu LK, Kesse-Guyot E et al. (2020) Ultraprocessed food consumption and risk of type 2 diabetes among participants of the NutriNet-Santé Prospective Cohort. JAMA Intern Med 180(2):283–291. 10.1001/jamainternmed.2019.5942 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Levy RB, Rauber F, Chang K et al. (2021) Ultra-processed food consumption and type 2 diabetes incidence: a prospective cohort study. Clin Nutr 40(5):3608–3614. 10.1016/j.clnu.2020.12.018 [DOI] [PubMed] [Google Scholar]
- 10.Duan MJ, Vinke PC, Navis G, Corpeleijn E, Dekker LH (2022) 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 20(1):7. 10.1186/s12916-021-02200-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Chen Z, Khandpur N, Desjardins C et al. (2023) Ultra-processed food consumption and risk of type 2 diabetes: three large prospective U.S. cohort studies. Diabetes Care 46(7):1335–1344. 10.2337/dc22-1993 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Martínez Steele E, Baraldi LG, Louzada ML, Moubarac JC, Mozaffarian D, Monteiro CA (2016) Ultra-processed foods and added sugars in the US diet: evidence from a nationally representative cross-sectional study. BMJ Open 6(3):e009892. 10.1136/bmjopen-2015-009892 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Wright JD, Folsom AR, Coresh J et al. (2021) The ARIC (Atherosclerosis Risk In Communities) Study: JACC Focus Seminar 3/8. J Am Coll Cardiol 77(23):2939–2959. 10.1016/j.jacc.2021.04.035 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Willett WC, Sampson L, Stampfer MJ et al. (1985) Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 122(1):51–65. 10.1093/oxfordjournals.aje.a114086 [DOI] [PubMed] [Google Scholar]
- 15.Shimakawa T, Sorlie P, Carpenter MA et al. (1994) Dietary intake patterns and sociodemographic factors in the atherosclerosis risk in communities study. ARIC Study Investigators. Prev Med 23(6):769–780. 10.1006/pmed.1994.1133 [DOI] [PubMed] [Google Scholar]
- 16.Monteiro CA, Cannon G, Moubarac JC, Levy RB, Louzada MLC, Jaime PC (2018) The UN Decade of Nutrition, the NOVA food classification and the trouble with ultra-processing. Public Health Nutr 21(1):5–17. 10.1017/S1368980017000234 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Willett WC, Howe GR, Kushi LH (1997) Adjustment for total energy intake in epidemiologic studies. Am J Clin Nutr 65(4 Suppl):1220S–1228S; discussion 1229S–1231S. 10.1093/ajcn/65.4.1220S [DOI] [PubMed] [Google Scholar]
- 18.Schneider ALC, Pankow JS, Heiss G, Selvin E (2012) Validity and reliability of self-reported diabetes in the Atherosclerosis Risk in Communities Study. Am J Epidemiol 176(8):738–743. 10.1093/aje/kws156 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Richardson MT, Ainsworth BE, Wu HC, Jacobs DR, Leon AS (1995) Ability of the Atherosclerosis Risk in Communities (ARIC)/Baecke Questionnaire to assess leisure-time physical activity. Int J Epidemiol 24(4):685–693. 10.1093/ije/24.4.685 [DOI] [PubMed] [Google Scholar]
- 20.Inker LA, Eneanya ND, Coresh J et al. (2021) New creatinine- and cystatin c-based equations to estimate GFR without race. N Engl J Med 385(19):1737–1749. 10.1056/NEJMoa2102953 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Harrell FE (2013) Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis. Springer Science & Business Media, New York [Google Scholar]
- 22.Chiuve SE, Fung TT, Rimm EB et al. (2012) Alternative dietary indices both strongly predict risk of chronic disease. J Nutr 142(6):1009–1018. 10.3945/jn.111.157222 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Poti JM, Braga B, Qin B (2017) Ultra-processed food intake and obesity: what really matters for health – processing or nutrient content? Curr Obes Rep 6(4):420–431. 10.1007/s13679-017-0285-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Valicente VM, Peng CH, Pacheco KN et al. (2023) ultraprocessed foods and obesity risk: a critical review of reported mechanisms. Adv Nutr 14(4):718–738. 10.1016/j.advnut.2023.04.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Martínez Steele E, Popkin BM, Swinburn B, Monteiro CA (2017) The share of ultra-processed foods and the overall nutritional quality of diets in the US: evidence from a nationally representative cross-sectional study. Popul Health Metr 15(1):6. 10.1186/s12963-017-0119-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.DiNicolantonio JJ, O’Keefe JH, Lucan SC (2015) Added fructose: a principal driver of type 2 diabetes mellitus and its consequences. Mayo Clin Proc 90(3):372–381. 10.1016/j.mayocp.2014.12.019 [DOI] [PubMed] [Google Scholar]
- 27.Risérus U, Willett WC, Hu FB (2009) Dietary fats and prevention of type 2 diabetes. Prog Lipid Res 48(1):44–51. 10.1016/j.plipres.2008.10.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Luukkonen PK, Sädevirta S, Zhou Y et al. (2018) Saturated fat is more metabolically harmful for the human liver than unsaturated fat or simple sugars. Diabetes Care 41(8):1732–1739. 10.2337/dc18-0071 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.McRae MP (2018) Dietary fiber intake and type 2 diabetes mellitus: an umbrella review of meta-analyses. J Chiropr Med 17(1):44–53. 10.1016/j.jcm.2017.11.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Tirosh A, Calay ES, Tuncman G et al. (2019) The short-chain fatty acid propionate increases glucagon and FABP4 production, impairing insulin action in mice and humans. Sci Transl Med 11(489):eaav0120. 10.1126/scitranslmed.aav0120 [DOI] [PubMed] [Google Scholar]
- 31.Debras C, Deschasaux-Tanguy M, Chazelas E et al. (2023) artificial sweeteners and risk of type 2 diabetes in the prospective NutriNet-Santé cohort. Diabetes Care 46(9):1681–1690. 10.2337/dc23-0206 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Walls KM, Hong KU, Hein DW (2023) Heterocyclic amines reduce insulin-induced AKT phosphorylation and induce gluconeogenic gene expression in human hepatocytes. Arch Toxicol 97(6):1613–1626. 10.1007/s00204-023-03488-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Zhang H, Ben Y, Han Y, Zhang Y, Li Y, Chen X (2022) Phthalate exposure and risk of diabetes mellitus: implications from a systematic review and meta-analysis. Environ Res 204(Pt B):112109. 10.1016/j.envres.2021.112109 [DOI] [PubMed] [Google Scholar]
- 34.Cardenas A, Hivert MF, Gold DR et al. (2019) Associations of perfluoroalkyl and polyfluoroalkyl substances with incident diabetes and microvascular disease. Diabetes Care 42(9):1824–1832. 10.2337/dc18-2254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Imamura F, O’Connor L, Ye Z et al. (2015) Consumption of sugar sweetened beverages, artificially sweetened beverages, and fruit juice and incidence of type 2 diabetes: systematic review, meta-analysis, and estimation of population attributable fraction. BMJ 351:h3576. 10.1136/bmj.h3576 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Greenwood DC, Threapleton DE, Evans CEL et al. (2013) Glycemic index, glycemic load, carbohydrates, and type 2 diabetes. Diabetes Care 36(12):4166–4171. 10.2337/dc13-0325 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Srour B, Chazelas E, Druesne-Pecollo N et al. (2023) Dietary exposure to nitrites and nitrates in association with type 2 diabetes risk: results from the NutriNet-Santé population-based cohort study. PLOS Medicine 20(1):e1004149. 10.1371/journal.pmed.1004149 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Zelber-Sagi S, Ivancovsky-Wajcman D, FlissIsakov N et al. (2018) High red and processed meat consumption is associated with non-alcoholic fatty liver disease and insulin resistance. J Hepatol 68(6):1239–1246. 10.1016/j.jhep.2018.01.015 [DOI] [PubMed] [Google Scholar]
- 39.Schwingshackl L, Hoffmann G, Lampousi AM et al. (2017) Food groups and risk of type 2 diabetes mellitus: a systematic review and meta-analysis of prospective studies. Eur J Epidemiol 32(5):363–375. 10.1007/s10654-017-0246-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Chen M, Sun Q, Giovannucci E et al. (2014) Dairy consumption and risk of type 2 diabetes: 3 cohorts of US adults and an updated meta-analysis. BMC Med 12:215. 10.1186/s12916-014-0215-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Choi HK, Willett WC, Stampfer MJ, Rimm E, Hu FB (2005) Dairy consumption and risk of type 2 diabetes mellitus in men: a prospective study. Arch Int Med 165(9):997–1003. 10.1001/archinte.165.9.997 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data described in the manuscript are available upon request pending application and approval from the National Heart, Lung, and Blood Institute (NHLBI) Biologic Specimen and Data Repository Information Coordinating Center (BioLINCC) (https://biolincc.nhlbi.nih.gov/home/) or the Atherosclerosis Risk in Communities study.