Ultra-Processed Food, Sleep and Type 2 Diabetes Risk

A prospective cohort analysis of 56,964 UK Biobank participants, published online ahead of print in the American Diabetes Association journal Diabetes on September 17, 2026, found that every 10% increase in ultra-processed food (UPF) consumption was associated with a 7% higher relative risk of developing incident type 2 diabetes. The researchers also reported that wearable-measured physical activity and adequate sleep were associated with lower risk across levels of genetic risk, including among people with high ultra-processed food intake.

Key takeaways:

• On September 17, 2026, researchers published prospective UK Biobank findings in Diabetes indicating that each 10% increment in UPF intake correlated with a 7% higher risk of type 2 diabetes (HR 1.07 [95% CI 1.03, 1.12]).
• Accelerometer data showed that sufficient moderate to vigorous physical activity (MVPA) and adequate sleep correlated with lower risk even among participants with high genetic susceptibility and high UPF intake.
• Because the investigation was an observational cohort study, the researchers cannot prove that industrial food processing causes type 2 diabetes, nor that exercise or sleep offsets ultra-processed food.
• Dietary intake was evaluated through the Nova processing classification system, which categorizes foods by industrial processing level rather than isolating single chemical additives or individual food brands.

What happened

On September 17, 2026, an observational study was published online ahead of print in Diabetes, a peer-reviewed journal of the American Diabetes Association. The paper, titled "Associations of Ultraprocessed Food Intake, Wearable-Measured Lifestyle Factors, and Genetic Susceptibility With Incident Type 2 Diabetes" (PubMed abstract, doi:10.2337/db26-0395), investigated how dietary processing interacts with objective lifestyle measures and inherited genetic risk factors.

The authors, Wenxin Xu, Tong Chen, Wei Zhang, Collin Sakal and Xinyue Li, are in the Department of Data Science, College of Computing, at City University of Hong Kong; Sakal is also affiliated with City University of Hong Kong (Dongguan). Europe PMC indexing indicates that the project received grant support from the Institute of Digital Medicine at the City University of Hong Kong. The full funding and conflict-of-interest statements are behind the journal's paywall.

The study analyzed records from 56,964 adult participants enrolled in the UK Biobank prospective cohort who were free of diagnosed diabetes at baseline. Diet was assessed with 24-hour dietary records, and physical activity and sleep were measured by accelerometers for seven consecutive days.

What changed

The study combined three kinds of data in one analysis: diet classified by the Nova processing system, activity and sleep measured by a wearable device, and a genetic risk score for type 2 diabetes, which the NIH's National Institute of Diabetes and Digestive and Kidney Diseases describes as the most common type of diabetes.

According to the published abstract, "Each 10% higher UPF intake was associated with 7% higher T2D risk (HR 1.07 [95% CI 1.03, 1.12])." Dose-response curve modeling showed a linear relationship between increased ultra-processed food consumption and incident type 2 diabetes. Conversely, moderate to vigorous physical activity (MVPA) demonstrated an L-shaped association (Pnonlinearity < 0.001), meaning risk was lower with more activity up to a point and then leveled off.

The authors evaluated inherited risk by building a polygenic risk score based on 424 single nucleotide polymorphisms associated with type 2 diabetes. Among individuals with high genetic susceptibility, lower incident risk was linked to lower UPF consumption (HR 0.65 [95% CI 0.49, 0.85]), higher total physical activity (HR 0.79 [95% CI 0.65, 0.97]), sufficient MVPA (HR 0.67 [95% CI 0.55, 0.81]), and adequate sleep (HR 0.57 [95% CI 0.39, 0.84]).

In joint analyses, sufficient MVPA (HR 0.69 [95% CI 0.50, 0.96]) and getting adequate sleep (HR 0.42 [95% CI 0.24, 0.75]) were linked to lower risk even among participants who had both high UPF consumption and high genetic risk scores. The authors wrote that wearable-derived evidence highlights MVPA and adequate sleep as behavioral targets for people with high UPF intake, and that lower UPF intake was associated with lower diabetes risk across all levels of genetic risk.

Who is affected

The study included 56,964 UK Biobank participants without diabetes at baseline who had dietary records and seven days of accelerometer data. The participants are UK adults, so the findings may not apply equally to every population.

The investigators grouped foods using the Nova classification framework, which categorizes food items into four tiers based on the nature, extent, and industrial purpose of processing rather than isolated nutrient metrics.

Nova's ultra-processed group is the one this study measured. Nova sorts foods by how they are processed, not by any single ingredient, so the paper says nothing about a particular product or additive.

What it means for shoppers

For grocery shoppers comparing options in supermarket aisles, this paper emphasizes overall dietary patterns rather than providing an assessment of any individual brand, product, or barcode. The study examined overall percentage intake of ultra-processed items across complete diets, meaning it did not analyze or isolate specific industrial food additives.

Shoppers seeking to recognize Nova group 4 products can inspect the ingredient list on packaged goods. Items containing industrial ingredients that are rare in home kitchens, such as flavorings, emulsifiers, sweeteners, colors and modified starches, are how Nova typically identifies ultra-processed foods. That is how the classification works, not a finding of this paper.

Consumers can check the Nutrition Facts panel for added sugars and dietary fiber while reviewing ingredient decks for long sequences of industrial additives.

These associations are not evidence that exercise or sleep cancels out ultra-processed food. The abstract says sufficient MVPA and adequate sleep were associated with lower risk among participants who had both high UPF intake and high genetic risk; it does not say they offset the diet.

The background

Controlled feeding trials have linked ultra-processed diets with weight gain, though what drives it is less clear, as covered in Ultra-Processed Food Trials: Weight Gain Clear, Cause Less So.

Other recent observational research has linked high ultra-processed food intake with Crohn's disease risk, as reviewed in Ultra-Processed Food, Crohn's Risk and Low DHA: New Study. Separating the effect of the food from other habits such as activity and sleep is one of the hard parts of this research.

The City University of Hong Kong team used seven days of accelerometer data and a 424-variant polygenic score in multivariable Cox models and restricted cubic splines to look at how activity, sleep and genetics relate to ultra-processed food intake and diabetes risk.

The study has clear limits. Because this was an observational study, it cannot demonstrate that ultra-processed foods directly cause type 2 diabetes, nor can it identify underlying biological mechanisms. Accelerometer tracking was captured during a single seven-day period, and diet came from 24-hour dietary records, so neither shows long-term habits.

The full paper is paywalled, and the abstract leaves out several details. The abstract does not disclose the total years of follow-up, the exact number of incident type 2 diabetes cases diagnosed, the numerical thresholds separating high from lower UPF consumption, the precise cutoff criteria for sufficient MVPA and adequate sleep, or the list of covariates adjusted for in the regression models. It also does not identify which specific ultra-processed food subcategories accounted for the statistical risk.

What to watch next

The full paper should give the follow-up length, the number of diabetes cases and the cut-offs behind high UPF intake, sufficient MVPA and adequate sleep, which the abstract does not report.

Observational studies like this one cannot show cause and effect; randomized trials are needed to test whether cutting ultra-processed food changes diabetes risk.

If you want to see what is in the packaged foods you buy, Aurascan is a free food-label scanner app in beta (iOS TestFlight, with a waitlist for other platforms). It pulls product data from Open Food Facts, USDA and barcode databases and explains every ingredient one by one with cited evidence, using per-ingredient verdicts (Clean, Mixed or Flagged) instead of a single score. Citations are checked by an automated verifier, and the app has no sponsors. Scan a product to read its ingredient list for yourself.