Intake of food additive preservatives and incidence of cancer: results from the NutriNet-Santé prospective cohort.

Introduction
Adding preservative additives to foods has become a standard practice in today’s food industry. In 2024, more than 20% of food items on the Open Food Facts World database contained at least one of these additives.1 The European parliament defines food additives as substances added to packaged products to prolong shelf life: protecting the foods against deterioration due to microorganisms, growth of pathogenic microorganisms, and deterioration as a result of oxidation, such as rancidity and colour changes.2

Between 2004 and 2025, the European Food Safety Authority (EFSA) re-evaluated 25 food preservative groups. This resulted in the establishment of reference values for acceptable daily intake of 16 preservatives or their respective groups.3 Acceptable daily intakes concerned a range of toxicological endpoints, including behavioural, carcinogenic, developmental, haematological, reproductive, and thyroid toxicity, as well as growth retardation, increased blood methaemoglobin levels, and increased mortality, all based on experimental data. A recent in vitro evaluation of the toxic effects of food additives in four human cell models suggested that some preservatives may have cytotoxic properties or enhance cell proliferation.4 It was suggested that several preservatives induce the production of advanced glycation end products5 and exert mutagenic5 and potentially carcinogenic activities.6
7
8 Some preservatives (such as ascorbic acid and alpha tocopherol) also occur in their natural forms in foods and beverages (eg, antioxidant vitamins C and E). Some epidemiological studies associated the consumption of these substances through natural dietary sources (eg, fruit, vegetables) with lower cancer risk.9 Such a beneficial property could be hypothesised to apply to the corresponding food additives. Other studies, however, raised the possibility that supplementation with these naturally occurring compounds may have harmful effects.10 Potentially, the same substance may involve different biochemical activities and have different health effects depending on the dose and the food matrix in which it is incorporated, with, for instance, the cell environment modulating a switch from antioxidant to pro-oxidant activity.11 Yet, these studies did not investigate food additives specifically. Except for rare preservatives such as nitrites and nitrates12
13
14
15
16 used in a limited number of products (ie, mostly processed meat), no data on intake of food preservatives were available in previous cohort studies owing to the lack of brand specific information and important variability in composition of the additives between commercial products. We therefore quantified the cumulative time dependent intake of preservatives and examined the associations with cancer incidence in a large prospective cohort with detailed dietary data.

Methods
This study followed the STROBE-NUT statement for observational studies in nutritional epidemiology.17 We used data from the French NutriNet-Santé prospective e-cohort, launched in 2009, to investigate the association between nutrition and health.18 The NutriNet-Santé protocol is available on the study website (https://info.etude-nutrinet-sante.fr/siteinfo/article/3). Participants aged 15 years and older were invited to participate in the study through a dedicated web based platform (https://etude-nutrinet-sante.fr/) and regularly answered questionnaires on dietary intakes, health, anthropometric measures,19
20 physical activity,21 lifestyle, and sociodemographic factors.22 Each participant provided an electronic informed consent form from the NutriNet-Santé cohort before enrolment.

Dietary data collection
At registration and every six months, participants completed series of three validated23
24
25 web based 24 hour dietary records. At each period, the dietary records were randomly assigned to three non-consecutive days over two weeks (two weekdays and one weekend day). Supplementary eMethod1 provides details on collection of dietary data and identification of individuals who reported unrealistically lower or higher energy intakes. Dietary intakes of energy, fibre, macronutrients, and micronutrients (including vitamins C and E) were assessed by merging with the NutriNet-Santé food composition table.26 Using multiple sources, we quantified participants’ intakes of naturally occurring acetic and citric acids, nitrites, nitrates, and sulfites (see supplementary eMethod2 for details).

Intakes of food additive preservatives
Assessment of food additive intake in the NutriNet-santé cohort through brand specific data from the 24 hour dietary records has been described previously (also see supplementary eMethod2).27 Briefly, we merged three composition databases with the NutriNet-Santé database to determine the presence of any specific food additives in industrial products. Dynamic matching was used to account for potential reformulations: products were matched on dates, and each participant’s date of consumption of each food or beverage was used to match the product to the closest composition data available, thus accounting for potential reformulations. Doses were determined by ad hoc laboratory analyses for the most frequently consumed additive-food pairs and doses retrieved from other sources such as the European Food Safety Authority after an official request for public access to the document. The 80 preservatives listed in the Codex General Standard for Food Additives database28 or UK Food Standards Agency29 were eligible for the present study. We decided to include preservatives in themselves as defined by regulation (EC) No 1333/20082 and antioxidants as both prevent the spoilage of food, with the latter preserving food through an antioxidant mode of action specifically. In this paper, we included preservatives with either non-antioxidant or antioxidant mode of action and all food additives with preservative properties. Some preservatives possessed additional key properties (eg, emulsification). We summed individual food preservatives with similar chemical structures into several groups: sorbates (European codes E200, E202, E203), benzoates (E210, E211, E212), sulfites (E220, E221, E222, E223, E224, E225, E228), nitrites (E249, E250), nitrates (E251, E252), acetates (E260, E261, E262, E263), propionates (E280, E281, E282), ascorbates (E300, E301, E302, E304), tocopherols (E306, E307, E307b, E307c), erythorbates (E315, E316), butylates (E319, E320, E321), citrates (E330, E332, E333), and EDTA (ethylenediaminetetra-acetate) (E385, E386).

Cancer ascertainment
We used a multi-source approach to ascertain participants with incident cancer. Throughout follow-up, participants could report health events, medical treatment, and examinations through the biannual health questionnaires or at any time directly through the health interface of their account. A physician expert committee validated every cancer event against official medical records. Moreover, the NutriNet-Santé cohort was linked to the national health insurance system database to collect additional information on medical treatments and consultations, and to the French national mortality registry to identify deaths and causes of death. Cases were then classified using ICD-10 (international classification of diseases, 10th revision) codes.30 In this study, we considered incident cancer to include all primary cancers diagnosed between enrolment and 31 December 2023, except for basal cell carcinoma of the skin.

Statistical analyses
Participants from the NutriNet-Santé cohort were included in the analysis if they completed at least two 24 hour dietary records during their first two years of follow-up, did not under-report or over-report energy intake, and did not have any prevalent cancer. Baseline participants’ characteristics are described as mean (standard deviation (SD)) for quantitative variables and as number (percentage) for qualitative variables for the overall population and according to each baseline sex specific third of total preservative intake. Baseline daily intakes of preservatives are reported as mean (SD) and median (interquartile range (IQR)), including per kilogram of body weight. We also computed the variations of preservative intake in participants with at least two intake periods, between the first and second half of follow-up (see supplementary eTable1). A correlation matrix was generated to visualise the Spearman correlations between intakes of individual additives (see supplementary eFigure1). For each studied preservative or group of preservatives, we categorised participants into lower, medium, and higher consumers defined as sex specific thirds if the preservative was consumed by at least a third of participants, and as non-consumers and lower or higher consumers separated by the sex specific median otherwise (cut-offs provided in supplementary eTable2). The relations between intake of preservatives coded as cumulative time dependent variables and cancer incidence were investigated using multivariable proportional hazard Cox models, with age as the time scale. Hazard ratios and 95% confidence intervals (CIs) were calculated. Participants contributed person time to the models from their age at enrolment in the cohort until age at the date of cancer diagnosis, death, last contact, or 31 December 2023, whichever occurred first. A counting process structure was used, with cumulative time dependent dietary variables updated every two years (preservative intake and dietary covariates). Intake during a specific period was computed using a weighted average of the most recent two year period and previous periods (see supplementary eMethod3), thereby using all available dietary record data. Based on the directed acyclic graph (see supplementary eFigure2), we adjusted the main model for sociodemographic (age (time scale), sex, educational level), anthropometric (body mass index, height), lifestyle and behavioural (physical activity, smoking status, number of smoked cigarettes in pack years), genetic predisposition (family history of cancer), and dietary factors (number of dietary records, time dependent daily intakes of energy without alcohol, alcohol, saturated fats, sodium, fibre, sugars, fruit and vegetables, dairy products, and red and processed meats—or haem iron for nitrites and nitrates models only). Breast cancer models were additionally adjusted for reproductive and hormonal factors: ever use of oral contraception (overall breast and premenopausal models only), age at menarche, number of biological children, menopausal status (overall breast model only), and hormonal treatment for menopause (overall breast and postmenopausal models only). In addition, when applicable we adjusted each model for intake of the corresponding substance from natural sources. We computed restricted cubic splines to investigate non-linear dose-response associations and tested the proportional hazard assumption using the Schoenfeld residual method. P values with and without correction for multiple testing by the false discovery rate were computed.31 Supplementary eMethods3 provides additional information and sensitivity analyses (mutual adjustment for preservatives other than the one studied; adjustment for percentage of total intake weight from ultra-processed foods; for intakes of colours, emulsifiers, and artificial sweeteners; for industrial trans fatty acid intake; for intakes of polyunsaturated fatty acids and haem iron; for standard deviation of individual participant intakes between periods, restricted to participants with at least two or three follow-up periods; and main model without update of intake cut-offs (defined at period 1)). Interactions between each preservative intake and percentage of ultra-processed foods in the diet (in weight), the number of dietary records per period, and antioxidants and smoking status were tested by entering the product of the two variables into Cox models.

Patient and public involvement
The research question in this article resulted from a strong concern of the participants involved in the NutriNet-Santé cohort, and of the public in general. Participants were not asked to advise on data interpretation or manuscript writing. This was primarily due to the lack of infrastructure, resources, funding, and time necessary to support public involvement in these aspects of the research process. However, participants provided feedback on our questionnaires and study design during follow-up, which we incorporated when feasible.

Results

Descriptive characteristics
The population study comprised 105 260 adults (fig 1), 78.7% of whom were women. On average, participants completed a total of 21 (SD 18) 24 hour dietary records. Mean age at baseline was 42.0 (SD 14.5) years. Compared with lower consumers (lowest third), higher consumers of total food preservatives (upper third) tended to be younger, less physically active, and less likely to have a family history of cancer, and they were likely to consume less alcohol and more ultra-processed foods and drinks (table 1). Table 2 and table 3 show the intakes of non-antioxidant and antioxidant preservatives, respectively. Overall, 99.7% of participants consumed food additive preservatives in the first two years of follow-up. Out of the 58 preservatives investigated in this study, 17 were consumed by at least 10% of the participants and thus were individually evaluated in association with cancer incidence (table 2 and table 3). In terms of proportion of consumers, the main preservatives were citric acid (91.7% of consumers), lecithins (87.1%), total sulfites (83.5%), ascorbic acid (83.5%), sodium nitrite (73.8%), potassium sorbate (65.5%), sodium erythorbate (52.7%), sodium ascorbate (50.3%), potassium metabisulfite (44.5%), and potassium nitrate (32.6%). No strong correlation between intakes of preservatives was identified (see supplementary eFigure1). Supplementary eTable1 displays the variations in preservative intake in participants with at least two intake periods, between the first and second half of follow-up. For all preservatives studied, most of the participants had stable intakes during the follow-up period, and when variations were observed, similar proportions of participants showed decreased and increased intakes.
Manufacturers used preservatives ubiquitously across various food groups (fig 2 and supplementary eTable3). Some were, however, more specific to given food groups—for example, 85.4% of sulfite intake was from alcoholic drinks; 53.8% of nitrites, 80.3% of nitrates, and 42.3% of erythorbates from processed meat (apart from processed red meat, offal, and poultry); 43.7% of propionates from refined grains and cereals and 30.4% from whole grains and cereals; 50.4% of ascorbates and 25.4% of citrates from processed fruit and vegetables; and 30.4% of tocopherols from breakfast cereals. In all, 34.6% of preservatives were consumed through ultra-processed foods in this population study (data not tabulated). For food additives that also occurred naturally in the diet, the relative contribution of the source of the food additive preservative varied depending on the compound: from on average 1% for tocopherols or 5% for acetates to 17% for citric acid, 29% for ascorbates, and 63% for sulfites.
No participants exceeded EFSA’s acceptable daily intake for sorbates, erythorbates, or nitrates.3 However, 90 participants exceeded the acceptable daily intake for sulfites, with a mean daily intake of 0.86 (SD 0.17) mg sulfur dioxide/kg body weight mg (median 0.81 (interquartile range 0.75-0.93) mg) and 54 exceeded the acceptable daily intake for nitrites, with a mean daily intake of 0.10 (0.03) mg nitrite ion/kg body weight (0.09 (0.08-0.10) mg).

Associations between preservative intake and cancer incidence
Participants’ mean follow-up was 7.57 (SD 4.56) years (796 944 person years). Between 2009 and 2023, 4226 participants received a diagnosis of incident overall cancer, comprising 1208 breast (387 premenopausal and 821 post-menopausal), 508 prostate, 352 colorectal, and 2158 other cancers. Schoenfeld residuals did not refute the proportional hazard assumption (see supplementary eFigure3).
Restricted cubic spline plots (see supplementary eFigure4) generally did not indicate a departure from linearity (P≥0.05 for non-linearity); P values for trend are provided below and in figure 3. For some associations, restricted cubic splines suggested a dose-response relation with plateau effect (P<0.05 for non-linearity)—in this case, the likelihood ratio overall P values (requiring no underlying hypothesis of linearity) are displayed thereafter and in forest plots. Both P trends and overall P values are provided for all tested preservatives in the main model (see supplementary eTable4) and in sensitivity analyses (see supplementary eTable5).
The results of Cox models are presented in figure 3 (non-antioxidants) and figure 4 (antioxidants) for overall (all sites), breast, and prostate cancers (see supplementary eTable4 for premenopausal and post-menopausal breast cancer and colorectal cancer models). No association was observed between total preservatives and cancer incidence (P=0.89 for overall cancer, P=0.95 for breast cancer, P=0.98 for prostate cancer, P=0.80 for colorectal cancer). Higher intakes of several preservatives were associated with higher cancer incidence: total non-antioxidants with overall cancer (P=0.001; absolute risks of cancer at age 60 years in higher consumers versus non-consumers or lower consumers 13.3%, 12.1%) and breast cancer (P=0.02; 5.7%, 4.8%); total sorbates with overall cancer (P=0.01; 13.1%, 12.3%) and breast cancer (P=0.02; 5.6%, 4.9%); potassium sorbate with overall cancer (P=0.01; 13.4%, 11.8%) and breast cancer (P=0.02; 5.7%, 4.6%); total sulfites with overall cancer (P=0.03; 13.4%, 11.9%); potassium metabisulfite with overall cancer (P=0.01; 13.5%, 12.0%) and breast cancer (P=0.01; 5.7%, 4.9%); total nitrites with overall cancer (P=0.004; 12.8%, 12.2%), prostate cancer (P=0.02; 4.3%, 3.4%), and breast cancer (P=0.003; 5.4%, 5.1%); sodium nitrite with overall cancer (P=0.003; 12.8%, 12.2%), prostate cancer (P=0.03; 4.2%, 3.4%), and breast cancer (P=0.002; 5.4%, 5.1%); total nitrates with overall cancer (P=0.001; 14.0%, 12.0%) and breast cancer (P=0.003; 5.9%, 4.8%); potassium nitrate with overall cancer (P=0.001; 14.0%, 12.0%) and breast cancer (P=0.003; 5.9%, 4.8%); total acetates with overall cancer (P=0.003; 14.3%, 12.2%) and breast cancer (P=0.02; 6.1%, 4.9%); acetic acid with overall cancer (P=0.01; 14.4%, 12.4%); total erythorbates with overall cancer (P<0.001; 13.5%, 11.9%) and breast cancer (P=0.01; 5.7%, 4.8%); and sodium erythorbate with overall cancer (P<0.001; 13.5%, 11.9%) and breast cancer (P=0.01; 5.7%, 4.8%). For colorectal cancer, the statistical power was limited owing to a lower number of participants with a diagnosis; no association was detected except for a suggested inverse relation with rosemary extracts (P=0.04; 1.0%, 1.2%).
Overall, associations were similar across all sensitivity analyses (see supplementary eTable4). Only a few exceptions were borderline with a restriction to participants with at least two two-year follow-up periods and lost statistical significance with ≥3 periods (eg, for total and sodium nitrites and overall and prostate cancers), probably owing to a reduced number of participants with incident cancer per person years (exclusion of more recently enrolled participants). No interaction was detected with percentage of ultra-processed foods in the diet (in weight), with number of dietary records in each period, or between antioxidants and smoking status (all P>0.05, data not tabulated).

Discussion
This study found multiple associations between preservatives that are widely used in industrial foods and beverages on the European market (potassium sorbate, potassium metabisulfite, sodium nitrite, potassium nitrate, acetic acid, and sodium erythorbate) and higher incidences of overall, breast, and prostate cancers. Most associations were observed for non-antioxidant preservatives. Among antioxidant preservatives, only total erythorbates and specific sodium erythorbate were found to be associated with higher incidence of cancer.

Comparison with other population studies
For some preservatives, EFSA data were available to compare intake levels with those observed in our population study (see supplementary eTable6). The order of magnitude was consistent overall. Relatively similar intakes were observed for sorbates, nitrates, propionates, ascorbates, lecithins, and extracts of rosemary. Compared with EFSA data, intakes tended to be lower in the NutriNet-Santé cohort for sulfites, nitrites, and alpha tocopherol, and higher for erythorbates. These differences may result from variations between methods for intake assessment (with data in the NutriNet-Santé cohort based on brand specific repeated 24 hour dietary records versus generic food items, and a generally lower number of records or recalls in studies on which EFSA estimates are based) and from differences in dates of assessment and study populations—for instance, the NutriNet-Santé cohort comprised more women and older participants with lower alcohol and processed meat intakes compared with the French general population.
According to the NOVA definition,32 food preservatives are not necessarily markers of ultra-processing (unlike other food additives, such as artificial sweeteners and colours). The proportion of additive preservatives from ultra-processed foods in this population study was 34.6%. This probably contributed to the results still being statistically significant after adjustment for the proportion of ultra-processed food in the diet.
That no other cohort study has investigated the associations between intakes of preservatives and cancer incidence probably relates to a lack of data on the consumption of specific industrial foods and thus great variations in additive content from one brand to another. Comparison with epidemiological literature is therefore not straightforward. Our group published a study on nitrites and nitrates and cancer incidence in NutriNet-Santé.12 Despite updated methodology (using time dependent cumulative intake) and longer follow-up (two additional years), the results remained globally stable, suggesting that higher intake of additive-originated nitrites was associated with higher incidence of prostate cancer, whereas intake of additive-originated nitrates was associated with higher incidence of breast cancer. A prospective investigation in the Netherlands Cohort Study found a positive association between higher intakes of nitrites from processed meat and pancreatic cancer.16 Another prospective study in the Iowa Women’s Health Study found a positive association between nitrites from processed meat and renal cancer among older women.15 Two other prospective studies reported positive associations between meat related nitrite and nitrate intakes and colorectal cancer.13
14 Consistently, the International Agency for Research on Cancer6 and French Food Safety Agency33 now recognise food additive nitrites and nitrates as probably carcinogenic if ingested under conditions that result in endogenous nitrosation, mainly for colorectal cancer. In the present study, the trend towards a positive association between colorectal cancer and nitrites (1.26 (0.95 to 1.68), P=0.1) and nitrates (1.23 (0.93 to 1.64), P=0.1) did not reach statistical significance, probably due to the limited number of incident colorectal cancers.
Some studies investigated workplace exposure to sulfites in factory workers, with mixed results.34
35 None, however, investigated dietary intake of food additives and cancer incidence.
A 2018 systematic review and dose-response meta-analysis of prospective studies found that although higher dietary intakes of vitamins C and E from natural sources and/or blood concentrations of vitamin C and alpha tocopherol were associated with reduced risk of total cancer, no evidence was available to support beneficial preventive effects of these antioxidants from other sources, and none of the included studies provided data specifically on food additives.9 In our study, we found no association between intakes of the corresponding food additives of those vitamins (E300, E301, E306, and E307) and cancer incidence. We were unable to find an epidemiological study that focused on food additive sources with which we could compare our results on other food preservatives.

Mechanistic plausibility
An in vitro study (24 hour treatment on four human cell models) on food additives suggested no cytotoxicity or genotoxicity for lecithins but cytotoxicity for potassium sorbate, sodium nitrite, sodium ascorbate, and sodium erythorbate, and enhanced cell proliferation for potassium metabisulfite, ascorbic acid, and citric acid.4 Potassium sorbate and sodium acetate could bind with serum albumin.5 Potassium sorbate might promote glycation of serum albumin, which is also associated with the production of advanced glycation end products.36 The compounds can alter immune and inflammatory pathways,2 potentially triggering immunosuppression and therefore the development of cancer. Additionally, sodium acetate (≤12.5 mM) has been reported to stimulate proliferation in human gastric adenocarcinoma cells and increase levels of interleukin 1β, interleukin 8, and tumour necrosis factor (TNF alpha) protein and mRNAs; therefore, potentially inflammatory conditions associated with cancer development.37 In another experimental study, high doses of potassium sorbate led to chromosomal aberrations in human blood lymphocytes.38 Potassium metabisulfite showed no carcinogenicity in mice models39 but enhanced gastric carcinogenicity in one rat study.40 We previously discussed the mechanistic plausibility of nitrites and nitrates.12 Briefly, hypotheses rely on the stepwise conversion of nitrate into nitrite in the body, followed by the formation of N-nitroso compounds, which according to the International Agency for Research on Cancer are probably carcinogenic to humans.6 Mechanistic studies suggest that the presence of nitrosatable compounds found in meat accelerate the formation of these N-nitroso compounds, whereas vitamin C and other antioxidants often found in fruit and vegetables inhibit it.6
41 In our models, associations between nitrites and nitrates and cancer incidences persisted after adjustment for intakes of haem iron, polyunsaturated fatty acids, and vitamins C and E. Vitamin C activity can switch from an antioxidant to pro-oxidant mode of action depending on the cell environment.11 We did not observe associations for ascorbic acid itself, but we did observe higher incidences of overall and breast cancers associated with erythorbates, which are isomers of ascorbates (but which do not impact vitamin C blood status3). Moreover, dosages in beverages in the USA have shown that benzene, a carcinogen to humans, may form at a nanogram per gram level when both benzoate salts and ascorbic or erythorbic acids are present.7 The result for an inverse association between rosemary extract and incidence of colorectal cancer in the present study should be considered with caution, given the limited number of participants with incident colorectal cancer. Rosemary contains phenolic compounds, including the two main molecules from which the food additive E392 is made (ie, carnosic acid and carnosol); polyphenols may modulate oxidative stress, cell growth, and cell differentiation thus potentially interfering with tumour development and progression.42 A rat model study suggested a chemopreventive action of rosemary extract for experimental mammary tumorigenesis.43 More generally, data are currently lacking, but it could be hypothesised that certain preservatives (some of which have antimicrobial properties) have an impact on gut microbiome and intestinal permeability (which in turn impacts on immune mechanisms, notably T regulatory cells44), as it has been shown for other type of additives (eg, some sweeteners and emulsifiers45
46
47). Additional studies are needed to better understand mechanisms linking food preservatives to the development of cancer, the differences in susceptibility to cancer location, and why erythorbates were the only preservative antioxidant associated with higher cancer incidence in this study. In the meantime, caution is needed in claiming innocuity of other antioxidant preservatives than erythorbates based on this study only, especially since this study focused on cancer and does not exclude potential effects on other health outcomes (eg, cardiometabolic health).

Strengths and limitations of this study
This prospective epidemiological study was based on a large cohort with highly detailed brand specific 24 hour dietary records for 14 years of follow-up (enabling assessment of time dependent cumulative intake). The cohort was linked to multiple food composition databases, ad hoc laboratory assays in the most frequently consumed additive-food pairs in food products, and dynamic matching accounting for reformulations, thereby providing access to unique information on food preservative intakes.
However, several limitations should be acknowledged. Firstly, the observational design does not allow for causality of the studied associations to be assumed based on this study alone. Residual confounding cannot be fully ruled out. The multivariable models were, however, adjusted for a broad spectrum of possible sociodemographic, anthropometric, lifestyle, and dietary confounding factors, limiting this potential bias. In particular, the food vectors of preservatives were diverse, encompassing variations in nutritional composition, which limited the risk of systematic bias by a same type of (poor) nutritional profile. For instance, for potassium sorbate, 26.3% of the intake was from fruit based and vegetable based products, whereas 21.6% was from fats and sauces. To limit confounding bias linked to nutritional profiles of vector foods as much as possible, we adjusted all models for energy, saturated fats, sodium, dietary fibre, and sugar intakes. Besides, mechanistic data from in vivo and in vitro studies support a potential causal involvement of these additives in carcinogenesis. Secondly, the generalisability of these results, collected though a web based study, should be considered. Validation studies comparing the NutriNet-Santé online dietary record versus interview with a trained dietitian25 and versus blood and urinary biomarkers of nutrient intakes,23
24 found that web based dietary studies appear as efficient and strategic tools for the collection of extensive and detailed information on dietary intakes for nutrition research.48
49 Integrated automated controls and pop-up warning messages contribute to limit errors (eg, aberrant food quantities, food omissions). It has also been suggested that the use of the internet reduces social desirability bias,50 which may enhance the quality of the data collected but also provide access to populations more difficult to reach otherwise. Nearly 95% of the French population has access to the internet,51 and we have shown that the study population was not limited to digitally fluent individuals.52 As in other studies investigating health and diet in which people enrol voluntarily, this study included more women, with a higher educational level and healthier lifestyles than the general French population.53
54 However, daily energy intake as well as proportion of energy from ultra-processed foods were similar in our population study compared with estimates from French nationally representative surveys, supporting the generalisability of our findings.55
56 Overall, the geographical distribution of the cohort also matched that of the general population in mainland France.57 Thirdly, although the assessment of intake was highly detailed, classification bias can never be totally excluded. For instance, using the Australian Food Composition Database to estimate naturally occurring acetic and citric acids was not optimal since variations between countries may occur, but French or European composition tables were not available or were less complete for these natural sources. Similarly, quantifying the intakes of the natural form of substances that also exist as food additive preservatives was impossible for some of them owing to limited data (eg, natural lecithins). Fourthly, it was not possible to investigate the association of several infrequently consumed preservatives with cancer (eg, benzoates). However, these limited proportions of consumers reflected a low occurrence on the French market, thus less potential for an impact of these substances on public health. Similarly, statistical power was limited for cancer locations other than the two most frequent in France (ie, breast and prostate), which may have limited our ability to detect associations for colorectal cancer in this study. Fifthly, the latency period between exposure to a carcinogen and the development of cancer varied from a short time to many years. Bioactive compounds may participate in the initiation of cancer but also trigger the development of pre-existing latent tumours due to other risk factors. Several large randomised controlled trials testing the impact of vitamin or mineral supplementation on cancer risk found that dietary factors can affect cancer risk with durations of use comparable to those in our study (eg, eight years for the SU.Vi.MAX trial58). Besides, associations were similar when restricting our population study to participants with longer follow-up. It will be interesting to re-run these analyses in the future to investigate longer term effects. Additional epidemiological and experimental studies are needed to better comprehend how food preservatives interact between themselves and with other food additives and food chemicals. Lastly, results were presented with and without adjustment for multiple testing by the false discovery rate method, with mostly stable results. Adjusting for multiple testing decreases type I error but also increases type II error (risk of false negative) and may lead to missing existing associations, which is why this adjustment is debated.59 Our results were, however, supported by mechanistic plausibility, and hazard ratios for most detected associations in our study were >1 (except for rosemary extract and colorectal cancer), strongly suggesting the associations were not going in random directions (which would have been the case if they were due to chance).

Conclusions
This large prospective cohort showed multiple positive associations between intake of widely consumed preservatives and increased incidence of overall, breast, and prostate cancers. These findings may have important public health implications given the ubiquitous use of these additives in a wide range of foods and beverages. Although replication in other epidemiological cohorts as well as additional in vivo and in vitro studies and short term trials are needed to better understand underlying mechanisms, these results are consistent with existing experimental data suggesting adverse cancer related effects of several of these compounds. This study brings new insights for the future re-evaluation of the safety of these food additives by health agencies, considering the balance between benefit and risk for food preservation and cancer. In the meantime, these results should encourage manufacturers to limit the use of unnecessary preservatives. Public health policies should be strengthened to promote and make accessible and affordable fresh, seasonal, homemade products to consumers, or even canned and other industrial foods, although minimally processed, that limit the use of preservatives and superfluous additives. Health professionals (general practitioners, dietitians) could play a key role in conveying these prevention recommendations to their patients.

What is already known on this topic
Preservatives are substances added to packaged foods to prolong shelf life, protecting against deterioration caused by micro-organisms and oxidation

Experimental in vivo and in vitro studies suggested negative impacts of preservatives through mechanisms involving advanced glycation end products, as well as mutagenic and potentially carcinogenic activities

What this study adds
Multiple positive associations between intake of preservatives widely used in industrial foods and higher cancer incidence (overall, breast, prostate) were observed in this large prospective cohort

If confirmed, these new data call for the re-evaluation of regulations governing the use of these additives by the food industry, to improve consumer protection and support recommendations for consumers to favour freshly made, minimally processed foods