Ultra-processed food and diabetes risk, tested outside a Western diet

Nearly all of the large cohort evidence linking ultra-processed food to metabolic disease comes from Western populations, whose diets and disease patterns differ from those elsewhere. A pooled analysis published in the European Journal of Epidemiology on September 16 tests the association in three prospective Korean cohorts, and then folds the result into an updated meta-analysis of 17 studies [s1].

The Korean association is present, positive, and small. The meta-analytic dose-response is tighter and more consistent than any single cohort.

What was measured

The analysis pooled individual-level data from three Korean prospective cohorts totalling 72,776 participants, with follow-up extending up to 18 years [s1]. Diet was assessed with validated food frequency questionnaires and classified by the Nova system, which sorts foods by the degree and purpose of industrial processing rather than by nutrient content [s1]. Intake was expressed as a percentage of grams per day — a unit choice that matters, and one that differs from the percentage-of-energy metric many Western analyses use [s1].

Incident type 2 diabetes was modelled with Cox proportional hazards regression [s1].

The findings

Participants in the highest quartile of ultra-processed food intake had a higher risk of type 2 diabetes than those in the lowest: pooled hazard ratio 1.11 (95% CI 1.02 to 1.21), with a test for trend at p = 0.03 [s1].

That association held after further adjustment for BMI, for nutritional factors including fibre, sodium, and carbohydrate intake, and for a diet quality score [s1]. This is the part of the result that carries most of the weight. A common objection to ultra-processed food research is that it is simply re-describing a diet high in sugar, salt, and fat, or a diet in people who are already heavier. Adjusting for diet quality and BMI does not fully answer that objection, but it narrows it.

Among individual food subgroups, processed meats, ready-to-eat or ready-to-heat mixed dishes, and ice cream were positively associated with diabetes risk [s1].

The meta-analysis of 17 prospective cohorts found that every 10% increment in ultra-processed food intake (in grams per day) was associated with a 10% higher risk, summary relative risk 1.10 (95% CI 1.08 to 1.12), in what the authors describe as a monotonic dose-response pattern [s1].

What this can and cannot establish

The confidence interval on the Korean estimate — 1.02 to 1.21 — excludes 1 but only just. A hazard ratio of 1.11 in an observational cohort is the size of effect that residual confounding can plausibly produce on its own. People who eat more ultra-processed food differ from those who eat less in ways that questionnaires capture imperfectly: shift work, income, cooking time, physical activity.

Food frequency questionnaires were not designed to classify processing. Nova assignment from an FFQ requires judgment about what a reported item actually was, and misclassification in that step is a known limitation across this literature.

The meta-analytic estimate is more precise, but precision across 17 observational cohorts is not the same as causal confidence. If all 17 share the same confounding structure — and cohorts of diet and disease largely do — pooling them narrows the interval without removing the bias.

The August 2025 American Heart Association science advisory on ultra-processed foods makes a related point from the other direction: mechanisms remain unclear, and most ultra-processed foods overlap with foods high in saturated fat, added sugars, and sodium that are already targets for cardiometabolic risk reduction [s2]. The advisory calls for identifying high-risk subgroups of ultra-processed foods rather than treating the category as uniform [s2], which is close to what the Korean subgroup analysis attempts.

Why testing outside Western cohorts matters

The composition of the ultra-processed category is not the same everywhere. What counts as an ultra-processed staple in Korea — including the ready-to-eat mixed dishes the analysis flags — differs from the sugar-sweetened beverages and packaged snacks that dominate the exposure in North American and European cohorts [s1].

If the association survives that substitution of contents, it is more likely to be about processing, or about something processing tracks, than about one particular set of products. If it did not survive, that would be equally informative. Here it survived, at modest magnitude.

What to watch

Whether trials that manipulate processing directly, rather than nutrients, reproduce a metabolic effect; whether the dose-response holds when intake is expressed in energy rather than grams; and whether regulatory definitions of ultra-processed food converge enough for the epidemiology to be compared across countries. The AHA advisory identifies exactly that definitional and regulatory work as a priority [s2].

This article describes observational cohort research and a meta-analysis of observational studies. It is informational only, is not medical advice, and does not recommend any diet.

Sources

• [s1] Kim Y, Cho Y, Koo B, Chen Z, Sun Q, Oh H, Ultra-processed food intake and risk of type 2 diabetes: a pooled analysis of three prospective cohorts of Korean adults and an updated meta-analysis, European Journal of Epidemiology, 2025;40:1293-1305, published online 2025-09-16.
• [s2] Vadiveloo MK, Gardner CD, Bleich SN, Khandpur N, Lichtenstein AH, Otten JJ, Rebholz CM, Singleton CR, Vos MB, Wang S, et al., Ultraprocessed Foods and Their Association With Cardiometabolic Health: Evidence, Gaps, and Opportunities: A Science Advisory From the American Heart Association, Circulation, 2025;152:e245-e263, published 2025-08-08.