A French cohort links several preservatives to cancer risk. Eleven of 17 showed nothing

Most research on processed food treats "ultra-processed" as a single category and asks whether eating more of it is associated with worse outcomes. A study published in The BMJ on 7 January took a different approach: it went after specific chemicals by name, and reported results for each one separately [s1].

That design produces a more useful result than a category-level association, and also a more awkward one. Several preservatives showed associations with cancer incidence. Most did not.

How the intakes were estimated

The NutriNet-Santé cohort is a French web-based study running since 2009. This analysis included 105,260 participants aged 15 or over who were free of cancer at baseline and had completed at least two 24-hour dietary records [s1]. Mean age was 42.0 (SD 14.5) years and 78.7% were women [s1].

Preservative intake was estimated as a cumulative, time-dependent exposure using repeated 24-hour records, cross-referenced against multiple food composition databases and supplemented with ad hoc laboratory assays on food products for the most frequently consumed additive-food pairings [s1]. That laboratory step is unusual and addresses a genuine weakness of additive research, which normally has to assume that a product contains what its category suggests.

Participants were grouped by intake — sex-specific thirds where at least a third of the cohort consumed the additive, otherwise non-consumers against lower and higher consumers split at the sex-specific median [s1]. Associations with cancer incidence were estimated with multivariable proportional hazards Cox models adjusted for potential confounders [s1].

Over a mean follow-up of 7.57 (SD 4.56) years, 4,226 participants received a cancer diagnosis: 1,208 breast, 508 prostate, 352 colorectal and 2,158 other [s1].

What was and was not associated

Higher intakes of several preservatives were associated with higher cancer incidence, comparing higher consumers with non-consumers or lower consumers [s1]:

• Total non-antioxidants: hazard ratio 1.16 (95% CI 1.07 to 1.26) for overall cancer, and 1.22 (1.05 to 1.41) for breast cancer [s1]
• Total sorbates, specifically potassium sorbate: 1.14 (1.04 to 1.24) overall, 1.26 (1.07 to 1.49) breast [s1]
• Total sulfites: 1.12 (1.02 to 1.24) overall [s1]
• Potassium metabisulfite: 1.11 (1.03 to 1.20) overall, 1.20 (1.04 to 1.38) breast [s1]
• Sodium nitrite: 1.32 (1.02 to 1.70) for prostate cancer [s1]
• Potassium nitrate: 1.13 (1.05 to 1.23) overall, 1.22 (1.05 to 1.41) breast [s1]
• Total acetates: 1.15 (1.06 to 1.25) overall, 1.25 (1.07 to 1.45) breast [s1]
• Acetic acid: 1.12 (1.01 to 1.25) overall [s1]
• Sodium erythorbate: 1.12 (1.04 to 1.22) overall, 1.21 (1.04 to 1.41) breast [s1]

The paper also reports these as absolute risks, which is more legible than hazard ratios. For total non-antioxidants, the absolute risk of cancer at age 60 was 13.3% in the higher-intake group against 12.1% in the comparison group [s1]. For breast cancer in the same comparison, 5.7% against 4.8% [s1]. The other associations sit in similar territory — differences of roughly one percentage point in absolute risk at age 60.

And the negative result, which the paper states plainly: 11 of the 17 individually studied preservatives were not associated with cancer incidence [s1].

Reading it honestly

This is a prospective observational cohort. It cannot establish that any of these additives causes cancer, and the authors do not claim it does. Their conclusion calls for epidemiology based on health effect biomarkers and experimental research to understand the pathways, and says that if confirmed, the findings call for re-evaluation of regulations governing the food industry's use of these additives [s1].

Three specific limits are worth holding on to. First, the cohort is heavily skewed — 78.7% women, recruited online, and volunteers for a nutrition study are not representative of a national population. Second, additives cluster: people who eat more potassium sorbate also eat more sulfites and more of everything else that comes in industrially processed food, and no statistical adjustment fully separates one chemical from the dietary pattern that carries it. Third, when a study tests 17 exposures against multiple cancer sites, some associations at p<0.05 will appear by chance, and the fact that 11 showed nothing is part of that picture rather than separate from it.

The authors' own practical framing is cautious and general: in the meantime, the findings support recommendations to favour freshly made, minimally processed foods [s1].

The regulatory argument alongside it

A review published in the Future Healthcare Journal on 22 January makes the policy case that the additives question is being neglected [s2]. Its authors note that ultra-processed foods now account for approximately half of caloric intake in many countries, notably among children, and that they are defined partly by the inclusion of non-nutritional additives — azo-dye colourants, non-caloric sweeteners, emulsifiers and preservatives [s2].

Their argument about why regulation has not followed the evidence has two parts. One is industry influence on political decisions, which they say there are increasing examples of in several countries [s2]. The other is more subtle: they contend that the prevailing policy focus on reducing fat, sugar and salt has distracted attention from the adverse effects of the alternative non-nutritional additives that get used instead [s2]. They propose changes to food labelling and public health interventions to reduce ultra-processed food consumption, with particular attention to socioeconomically disadvantaged populations and children [s2].

That review is an argument, not an experiment, and it acknowledges that association studies cannot account for all potential confounding — arguing instead that triangulation across cohort studies, animal and in vitro models and human challenge studies makes the case more compelling [s2].

What to watch

Whether European food safety regulators respond to additive-specific rather than category-specific evidence, and whether the NutriNet-Santé associations replicate in cohorts with different dietary patterns and a less skewed sex distribution.

Sources

1. [s1] Intake of food additive preservatives and incidence of cancer: results from the NutriNet-Santé prospective cohort. BMJ, 7 January 2026. https://doi.org/10.1136/bmj-2025-084917
2. [s2] Why is there no regulation despite evidence that ultra-processed foods are hazardous to long-term health? Future Healthcare Journal, 22 January 2026. https://doi.org/10.1016/j.fhj.2026.100503