Plant-Based Diets, Ultra-Processed Foods, and Risks of Mortality and Major Chronic Diseases: A Prospective Cohort Study

What the plant based diet index and Nova is

The plant based diet index scores a diet by how much of it comes from plant foods, with a healthy version that rewards whole grains, fruit, vegetables, nuts and legumes, and an unhealthy version that rewards refined grains, sweets and sugary drinks.

Nova is a separate system that sorts foods by how much industrial processing they have been through rather than by their nutrients, and its top category, ultra processed, is what the term refers to.

Why they ran it

Global guidelines push diets towards plant foods, yet many modern plant based products are industrially formulated meat and dairy alternatives that fall into the ultra processed category, and the authors note that vegetarians and vegans may consume more of these than omnivores.

They say that no study had yet examined whether ultra processed content changes the link between plant based dietary patterns and several major chronic diseases at once, while also accounting for nutrient quality.

Drawn from the paper's introduction and its research in context section.

This one dates from 5 June 2026 and we are catching up with it now. It followed 124,836 UK Biobank participants aged 40 to 70, recruited between 2006 and 2010, with a mean age of 56.2 years and 55.8 percent women. Diet came from a 24 hour online recall. The team built four versions of a plant based diet index, separating healthy from unhealthy plant based patterns and, within each, high from low ultra processed content, using the Nova classification and a modified nutrient quality index. Participants were tracked for 8.3 to 10.5 years using multivariable Cox models.

Over follow up there were 5780 deaths, 3420 type 2 diabetes cases, 6078 cardiovascular cases and 9437 cancers. Closer adherence to a healthy plant based pattern tracked with 8 to 28 percent lower risk across outcomes. For all cause mortality the hazard ratios were 0.92, interval 0.85 to 1.00, for the high ultra processed version and 0.91, interval 0.84 to 0.98, for the low. Diabetes showed the widest gap between the two: 0.89, interval 0.79 to 0.99, for high, and 0.72, interval 0.65 to 0.79, for low. Cardiovascular risk was 11 percent lower on the high ultra processed pattern.

The headline is that processing level did not erase the association, and nutrient quality came out similar across the high and low ultra processed healthy patterns. The detail is less tidy. The mortality interval for the high ultra processed pattern touches 1.00 at its upper bound. The diabetes advantage is far larger in the low ultra processed version. And the clearest cardiovascular signal shows up in the high ultra processed pattern, with no clear association for the low one. One sentence of conclusion is covering results that behave quite differently by outcome.

The numbers

undefined	undefined
undefined	undefined
undefined	undefined
undefined	undefined
undefined	undefined

Why this might happen

Proposed by the authors This is the explanation the authors offer in their discussion. This study did not test it.

The authors argue the pattern reflects nutrient quality rather than processing level, pointing out that the high and low ultra processed versions of the healthy plant based index carried similar nutrient quality scores, with more fibre and less saturated fat in both.

They also argue that ultra processed foods are heterogeneous, with the adverse signal concentrated in specific categories such as sugar sweetened beverages, while wholegrain rich ultra processed items tracked with lower risk in their food group analyses.

They acknowledge that non nutritive features of ultra processed foods have been linked to inflammation, oxidative stress, impaired satiety signalling and gut dysbiosis, but conclude that in their data those risks appear small next to overall diet quality.

Drawn from the paper's discussion.

What this does not show

It cannot show that eating this way causes lower risk. Cox models adjust only for the confounders that were measured and entered. People who score high on the healthy plant based index also tend to differ in smoking, activity, income, education and health care contact, and residual confounding along those lines can produce associations of exactly this size.
It cannot describe anyone's usual long term diet. Exposure came from a self completed 24 hour recall. Every index score, including the split between high and low ultra processed patterns, is built from what people remembered about a small number of days near recruitment.
It cannot tell an individual how much their own risk would change. A hazard ratio compares event rates between groups across 8.3 to 10.5 years of follow up. It is a relative measure, and the abstract reports no absolute risk differences.
It cannot show that ultra processed content is irrelevant. The comparison is between two ranked indices, not between matched diets differing only in processing. The diabetes hazard ratios, 0.89 for high against 0.72 for low, differ substantially, and the mortality interval for the high pattern reaches 1.00.

Where this leaves us

Within this cohort it is now more likely that how healthful a plant based pattern is tracks outcomes more closely than whether its components are ultra processed. That contradicts an earlier French cohort result of no cardiovascular benefit for high ultra processed healthy plant based diets, and it refines the broader claim that ultra processed foods are uniformly harmful, which the authors say their findings challenge.

UK Biobank volunteers aged 40 to 70 recruited between 2006 and 2010, 96.6 percent of them White, who completed at least two 24 hour dietary assessments. The authors state this group is healthier, more educated and more health conscious than the underlying population.

The authors call for research in more diverse populations, with repeated dietary assessments capturing longer term trajectories and current product formulations, plus mechanistic work on different types of ultra processed foods.

Caveats worth holding

UK Biobank is a volunteer cohort recruited between 2006 and 2010, so it is not a cross section of the UK population and the event rates here should not be read as national rates.
Follow up of 8.3 to 10.5 years is short relative to how long chronic disease takes to develop, particularly for cancer.
Diet was captured through short online recalls near the start of follow up, while eating habits over a decade can move considerably.
The work was funded by Research Ireland, Northern Ireland's Department of Agriculture, Environment and Rural Affairs, and UK Research and Innovation, within a co-centre whose remit is sustainable food systems.