Four of nine ultra-processing marker categories tracked hypertension risk. Five did not.

The persistent weakness in the ultra-processed food literature is that "ultra-processed" is a category, not a mechanism. The NOVA classification groups foods by how industrially they were made, and cohort after cohort finds the top of that category associated with worse outcomes. What almost none of them can say is which part of the processing is doing it — or whether processing is doing anything at all, as opposed to the salt, sugar and fat that tend to travel with it.

A UK Biobank analysis published in the European Heart Journal on August 3 takes a step at that question by disaggregating the category [s1].

The method

The study followed 101,560 UK Biobank participants who were free of hypertension at baseline and had completed at least one 24-hour dietary recall using the Oxford WebQ instrument [s1].

Rather than scoring each food as ultra-processed or not, the researchers matched up to ten commercial products to every Oxford WebQ item and read the ingredient lists [s1]. That produced 37 specific markers of ultra-processing, grouped into nine categories [s1]. Intake of foods containing each marker was modelled as a percentage of total food intake, using penalised cubic splines, with Cox proportional hazards models for incident hypertension [s1].

Because the dose-response curves are not straight lines, hazard ratios are expressed relative to the curve's minimum — the "HR nadir," rescaled to 1 [s1].

Over a median 10.5 years of follow-up, 7,633 people developed hypertension [s1].

What separated

Four of the nine categories were positively associated with hypertension risk [s1]:

• Flavour — 40% versus 8% of total food intake: HR 1.37 (95% CI 1.22–1.54)
• Colouring agent — 20% versus 0%: HR 1.55 (95% CI 1.36–1.77)
• Sweetener — 20% versus 0%: HR 1.42 (95% CI 1.31–1.55)
• Varieties of sugar — 10% versus 0%: HR 1.19 (95% CI 1.11–1.27)

Five categories showed no significant association: flavour enhancer, processing aid, modified oil, protein source, and added fibre [s1].

Among the 37 individual markers, foods containing acesulfame, aspartame, erythritol, saccharin, sucralose, xylitol, bulking agent, dextrose, fructose, hydrolysed protein and whey protein were significantly and positively associated with hypertension risk [s1]. Foods containing firming agents, foaming agents and glazing agents showed negative correlations [s1].

How to read that

Carefully, and the authors label the study exploratory themselves [s1].

The exposure is not the additive. It is foods containing the additive. A food with a colouring agent in it is a different food in many other ways too, and the design cannot separate the marker from the product it identifies. The negative correlations for glazing and foaming agents are a useful reminder of this: no one is proposing that glazing agents lower blood pressure. They are markers of some other category of food.

Twenty-four-hour dietary recall is also a blunt instrument for a decade of eating, and the ingredient matching — up to ten commercial products per survey item — is a modelled approximation of what each participant actually consumed, not a record of it.

Residual confounding runs in the usual direction. People whose diets are heavy in brightly coloured sweetened products differ from people whose diets are not, in ways that no covariate list fully captures.

Why it is still worth something

Because the alternative — treating ultra-processed food as one undifferentiated block — has produced a literature that is broad and unactionable. Associations now span outcome after outcome: a separate UK Biobank cohort of 154,376 participants published in July reported that the highest quartile of ultra-processed food intake carried a 20% higher risk of cholecystitis (HR 1.20; 95% CI 1.07–1.34) and a 15% higher risk of cholelithiasis (HR 1.15; 95% CI 1.05–1.25) over a median 13.2 years [s2].

Findings like that accumulate without narrowing anything. A study that says four ingredient categories carry the signal and five do not is at least a testable claim, and the authors frame it that way: the four positive categories are named as prime candidates for reduction strategies and for mechanistic study [s1].

Mechanistic study is the operative phrase. Until a controlled feeding trial isolates one of these categories, this is a hypothesis generated from observational data — a sharper hypothesis than the field had, and still a hypothesis.

This article is informational and is not medical advice.

Sources

• Markers of ultra-processed food and incident arterial hypertension in the UK Biobank — European Heart Journal, 2026-08-03
• Ultra-processed food and risk of cholecystitis and cholelithiasis: a prospective cohort study — Journal of Health, Population and Nutrition, 2026-07-13