eAppendix. Details and Examples of the NOVA Classification
eTable 1. Sociodemographic and Lifestyle Characteristics of the Study Population According to Quartiles of Proportion of Ultra-processed Food in the Diet (n = 44,551)
eTable 2. Means of Proportion of Ultra-processed Food in the Diet of Adults From the Study Population (n = 44,551)
eTable 3. Sociodemographic and Lifestyle Characteristics of the Study Population According to Quartiles of the Energy-weighted Proportion of Ultra-processed Food in the Diet (n = 44,551)
eTable 4. Nutrient Intake According to Quartiles of Proportion of Ultra-processed Food in the Diet of Adults From the Study Population (n = 44,551)
eTable 5. Association Between Ultra-processed Food Consumption and Nutrient Intake of Adults From the Study Population (n = 44,551)
eTable 6. Nutrient Intake According to Quartiles of the Energy-weighted Proportion of Ultra-processed Food in the Diet of Adults From the Study Population (n = 44,551)
eTable 7. Association Between Quartiles of Proportion of Ultra-processed Food in the Diet and Overall Mortality Risk in Adults From the NutriNet-Santé Cohort (n = 44,551)
eTable 8. Association Between the Proportion of Ultra-processed Food in the Diet and Overall Mortality Risk in Adults From the NutriNet-Santé Cohort (Primary Analysis, Case Complete)
eTable 9. Association Between the Energy-weighted Proportion of Ultra-processed Food in the Diet and Overall Mortality Risk in Adults From the NutriNet-Santé Cohort (n = 44,551)
eTable 10. Association Between the Proportion of Ultra-processed Food in the Diet and Overall Mortality Risk in Adults from the NutriNet-Santé Cohort (Sensitivity Analyses, Case Complete)
eTable 11. Association Between the Proportion of Ultra-processed Food in the Diet and Overall Mortality Risk in Adults From the NutriNet-Santé Cohort (Sensitivity Analyses, Imputed Data)
Customize your JAMA Network experience by selecting one or more topics from the list below.
Schnabel L, Kesse-Guyot E, Allès B, et al. Association Between Ultraprocessed Food Consumption and Risk of Mortality Among Middle-aged Adults in France. JAMA Intern Med. 2019;179(4):490–498. doi:10.1001/jamainternmed.2018.7289
Is high consumption of ultraprocessed food associated with an increase in overall mortality risk?
In this cohort study of 44 551 French adults 45 years or older, a 10% increase in the proportion of ultraprocessed food consumption was statistically significantly associated with a 14% higher risk of all-cause mortality.
An increase in ultraprocessed food consumption may be associated with an overall higher mortality risk; further prospective research is needed to confirm these findings.
Growing evidence indicates that higher intake of ultraprocessed foods is associated with higher incidence of noncommunicable diseases. However, to date, the association between ultraprocessed foods consumption and mortality risk has never been investigated.
To assess the association between ultraprocessed foods consumption and all-cause mortality risk.
Design, Setting, and Participants
This observational prospective cohort study selected adults, 45 years or older, from the French NutriNet-Santé Study, an ongoing cohort study that launched on May 11, 2009, and performed a follow-up through December 15, 2017 (a median of 7.1 years). Participants were selected if they completed at least 1 set of 3 web-based 24-hour dietary records during their first 2 years of follow-up. Self-reported data were collected at baseline, including sociodemographic, lifestyle, physical activity, weight and height, and anthropometrics.
The ultraprocessed foods group (from the NOVA food classification system), characterized as ready-to-eat or -heat formulations made mostly from ingredients usually combined with additives. Proportion (in weight) of ultraprocessed foods in the diet was computed for each participant.
Main Outcomes and Measures
The association between proportion of ultraprocessed foods and overall mortality was the main outcome. Mean dietary intakes from all of the 24-hour dietary records available during the first 2 years of follow-up were calculated and considered as the baseline usual food-and-drink intakes. Mortality was assessed using CépiDC, the French national registry of specific mortality causes. Hazard ratios (HRs) and 95% CIs were determined for all-cause mortality, using multivariable Cox proportional hazards regression models, with age as the underlying time metric.
A total of 44 551 participants were included, of whom 32 549 (73.1%) were women, with a mean (SD) age at baseline of 56.7 (7.5) years. Ultraprocessed foods accounted for a mean (SD) proportion of 14.4% (7.6%) of the weight of total food consumed, corresponding to a mean (SD) proportion of 29.1% (10.9%) of total energy intake. Ultraprocessed foods consumption was associated with younger age (45-64 years, mean [SE] proportion of food in weight, 14.50% [0.04%]; P < .001), lower income (<€1200/mo, 15.58% [0.11%]; P < .001), lower educational level (no diploma or primary school, 15.50% [0.16%]; P < .001), living alone (15.02% [0.07%]; P < .001), higher body mass index (calculated as weight in kilograms divided by height in meters squared; ≥30, 15.98% [0.11%]; P < .001), and lower physical activity level (15.56% [0.08%]; P < .001). A total of 602 deaths (1.4%) occurred during follow-up. After adjustment for a range of confounding factors, an increase in the proportion of ultraprocessed foods consumed was associated with a higher risk of all-cause mortality (HR per 10% increment, 1.14; 95% CI, 1.04-1.27; P = .008).
Conclusions and Relevance
An increase in ultraprocessed foods consumption appears to be associated with an overall higher mortality risk among this adult population; further prospective studies are needed to confirm these findings and to disentangle the various mechanisms by which ultraprocessed foods may affect health.
Ultraprocessed foods are food products that contain multiple ingredients and are manufactured through a multitude of industrial processes.1 These food products are usually ready to heat and eat, affordable, and hyperpalatable. Examples include mass-produced and packaged snacks, sugary drinks, breads, confectioneries, ready-made meals, and processed meats.
Consumption of ultraprocessed foods has increased during the past several decades2-5 and is associated with an overall unbalanced nutritional profile.6 Prospective cohort studies assessed that ultraprocessed foods intake was associated with a higher risk of dyslipidemia,7 obesity,8 hypertension,9 and cancer.10 Nutritional characteristics of ultraprocessed foods could partly explain the development of noncommunicable chronic diseases among those who consume them. Ultraprocessed foods are generally energy dense, rich in refined carbohydrates, saturated fats, and salt, and contain low dietary fiber. These features have been associated with several noncommunicable diseases that are the leading causes of mortality.11,12 Beyond their nutritional aspects, ultraprocessed foods have specific characteristics, owing to the industrial processes they undergo. Thus, concern is rising about the potential harmful health consequences of neoformed contaminants13 or food additives.14,15
Cardiovascular disease (CVD), respiratory disease, cancer, and diabetes account for more than 80% of all premature deaths from noncommunicable disease worldwide.16 Associations between ultraprocessed foods intake and noncommunicable diseases have been documented,7-10 but no study has directly investigated the association between ultraprocessed foods consumption and mortality. Our aim was to assess the association between ultraprocessed foods consumption and overall mortality risk in a large cohort study of French adults 45 years or older.
The NutriNet-Santé Study was approved by the institutional review board of the French Institute for Health and Medical Research and the Commission Nationale de l’Informatique et des Libertés. All participants provided electronic informed consent. Institutional review board approval and informed consent for the NutriNet-Santé Study apply to the current study.
The current study selected participants from the NutriNet-Santé Study, an ongoing prospective observational cohort study that launched on May 11, 2009, and performed a follow-up through December 15, 2017. The NutriNet-Santé Study includes a large voluntary population of adults in France, and its details are described elsewhere.17
At the time of inclusion in the NutriNet-Santé Study, participants completed 5 questionnaires pertaining to their own sociodemographic data, anthropometrics, physical activity, and health status.18-21 Participants were invited to provide a set of 3 nonconsecutive, web-based 24-hour dietary records (records were randomly assigned over a 2-week period) every 6 months and to complete all 5 questionnaires every year.
The 24-hour dietary records were a detailed collection of food and beverages consumed at each of the 3 main meals daily and at any other eating occasions. Participants estimated the quantities they consumed using validated photographs of portion sizes.22,23 The NutriNet-Santé Study food composition table contains more than 3000 references reflecting the foods consumed by the French population. Detailed information on macronutrient and micronutrient composition of each food is available in a public database.24 These self-administered 24-hour dietary records were previously validated through comparisons with interviews by dietitians25 and biomarkers of nutritional status.26,27
Eligible participants in this current study had completed at least 1 food set during the first 2 years of follow-up. We identified dietary underreporting using the Black method and excluded underreporters of energy intake.28 We calculated the mean dietary intakes from all of the 24-hour dietary records available during the first 2 years of follow-up and considered these as the baseline usual dietary intakes.
Each of the 3000 foods in the NutriNet-Santé Study composition table was classified according to the NOVA food classification system, which categorizes food products into 4 groups according to the nature, extent, and purpose of processing.1 This current study focused on 1 group classified as ultraprocessed foods, which are manufactured industrially from multiple ingredients that usually include additives used for technological and/or cosmetic purposes. Ultraprocessed foods are mostly consumed in the form of snacks, desserts, or ready-to-eat or -heat meals. A comprehensive description of the NOVA food classification system is in eAppendix in the Supplement.
Self-reported data were collected at baseline, including sociodemographic, lifestyle, physical activity level measured with the International Physical Activity Questionnaire,21 and weight and height, which were used to compute body mass index (BMI). Self-reported anthropometrics have previously been validated when compared with measured data.19
Vital status of all participants and causes of death were systematically verified with CépiDC, the French national registry of specific causes of mortality. Causes of death from chronic diseases were identified using the International Classification of Diseases, Tenth Revision.29
Of the 158 361 participants in the NutriNet-Santé Study, 98 019 provided at least three 24-hour dietary records within the first 2 years of follow-up. We excluded participants younger than 45 years because nutrition-associated causes of deaths were less likely in younger participants. To perform a prospective analysis, we excluded participants who died within the first year of follow-up. Thus, 44 551 people were included for analysis (Figure). We used multiple imputations (the fully conditional specification method) to replace missing values.
The proportion in weight of ultraprocessed foods in the diet (% grams per day) was computed for each participant. The energy proportion of ultraprocessed foods in the diet (% kilocalorie per day) was compared with the ultraprocessed foods proportion in a previous study,6 and proportion in weight was the preferred measure to better consider ultraprocessed foods that did not contribute to energy intake and had features directly related to food processing. For comparison with international data, we also calculated the energy proportion of ultraprocessed foods in the diet.
Adherence to French nutritional recommendations was assessed using the modified Programme National Nutrition Santé Guideline Score (PNNS-GS; no minimum score; maximum: 13.5, with the highest score indicating better compliance with the national nutritional recommendations); this validated score has been described previously.30 In this current study, using the modified PNNS-GS, we excluded physical activity and considered food-related components only.31
Unadjusted means of the proportion of ultraprocessed foods in the diet were calculated and presented for descriptive statistics of the population, using appropriate unpaired, 2-tailed t tests or analysis of variance for assessing the between-group differences. Macronutrient and micronutrient intake were log-transformed to approximate normality (log-nutrient). Multivariable regression analysis was performed to assess the association between log-nutrient intake and weight proportion, adjusted for sex, age, and total energy intake. We estimated the proportion of change in nutrient intake for a 10% increase in ultraprocessed foods proportion, using the following transformation of the β coefficient: 100 × (e(β) − 1). These analyses were replicated using the energy proportion of ultraprocessed foods in the diet. Analyses were replicated using proportions in weight expressed as quartiles.
We used Cox proportional hazards regression models to estimate the association between ultraprocessed foods proportion (coded as a continuous variable and as quartiles) and overall mortality. Participants contributed person-time until the date of death, the date of last news, or December 15, 2017, whichever occurred first. We estimated hazard ratios (HRs) and 95% CIs for a 10% increase in ultraprocessed foods proportion.
We ran 3 models that were incrementally adjusted to account for known or potential risk factors and confounding factors. Models were computed both on complete cases and imputed data (10 samples). The 3 models were as follows: model 1, minimally adjusted for sex (with age as timescale); model 2, model 1 plus monthly income per household unit (<€1200, €1200-€1799, €1800-€2299, or ≥€2300), educational level (no diploma or primary school, secondary school, graduate studies ≤3 years, or graduate studies >3 years), marital status (married or cohabiting, or single, divorced, widowed), residence (rural, urban unit <20 000 inhabitants, urban unit 20 000-200 000 inhabitants, or urban unit >200 000 inhabitants), BMI (continuous; calculated as weight in kilograms divided by height in meters squared), physical activity level (low, moderate, or high), smoking status (never smoked, former smoker, or current smoker), alcohol intake (continuous; grams per day), energy intake (without alcohol, continuous; kilocalorie per day), first-degree family history of cancer (yes or no), first-degree family history of CVDs (yes or no), number of 24-hour dietary records (continuous), and season of dietary records; and model 3, model 2 plus modified PNNS-GS (continuous). Separating models 2 and 3 enabled us to consider the association between ultraprocessed foods and mortality risk independently from the overall nutritional quality of diet. P values for trend were computed using quartiles as ordinal variables. We confirmed that the proportional hazards assumption was satisfied through the examination of the Schoenfeld residual plots.
Two-factor interactions were tested on model 3 between ultraprocessed foods consumption and physical activity, smoking status, and BMI. Because no substantial interaction was detected, all analyses were performed in the overall sample.
We conducted 5 sensitivity analyses by (1) excluding cases of deaths occurring during the first 2 years of follow-up; (2) testing further adjustments on model 3 for prevalent CVDs (myocardial infarction, acute coronary syndrome, stroke, and angioplasty), risk factors for CVDs (type 2 diabetes; hypertension, and dyslipidemia), and prevalent cancer; (3) testing further adjustment on model 3 for a Western dietary pattern derived from principal components analysis; (4) testing the association between ultraprocessed foods and mortality with the absolute amount of ultraprocessed foods consumed (in weight) as the main exposure; and (5) excluding prevalent CVDs and cancers at baseline.
All tests were 2-sided, and a 2-sided P < .05 was considered statistically significant. Data management and statistical analysis were performed using SAS software, version 9.4 (SAS Institute Inc).
A total of 44 551 participants were included, of whom 32 549 (73.1%) were women, with a mean (SD) age at baseline of 56.7 (7.5) years. The mean (SD) number of dietary records during the first 2 years of follow-up was 6.6 (3.1).
Consumption of ultraprocessed foods represented a mean (SD) of 14.4% (7.6%) of the total weight of food and drink intake, corresponding to a mean (SD) of 29.1% (10.9%) of total energy intake. Higher consumption of ultraprocessed foods was associated with younger age (45-64 years, mean [SE] proportion of food in weight, 14.50% [0.04%]; P < .001), lower income (<€1200/mo, 15.58% [0.11%]; P < .001), lower educational level (no diploma or primary school, 15.50% [0.16%]; P < .001), living alone (15.02% [0.07%]; P < .001), higher BMI (≥30, 15.98% [0.11%]; P < .001), and lower physical activity level (15.56% [0.08%]; P < .001) (Table 1). Similar results were obtained with ultraprocessed foods proportion coded as quartiles and with energy proportion (eTables 1-3 in the Supplement).
Higher ultraprocessed foods consumption was associated with a lower adherence rate to nutritional recommendations, with a lower modified PNNS-GS (−5.7% for a 10% increase in ultraprocessed foods proportion; P < .001). A 10% increment in ultraprocessed foods proportion was associated with higher intake of saturated fats (change in nutrient intake, 3.05% [0.002%]; P < .001), and sugars (3.59% [0.002%]; P < .001). Conversely, a 10% increment in ultraprocessed foods proportion was associated with lower intake of fiber (−4.88% [0.002%]; P < .001; Table 2). Similar results were obtained analyzing ultraprocessed foods proportion coded as quartiles and energy proportion (eTables 4-6 in the Supplement).
During follow-up, 602 deaths (1.4%) occurred, including 219 deaths caused by cancer and 34 by CVDs. The median (interquartile range) follow-up time was 7.1 (5.7-7.9) years (mean [SD] follow-up, 6.6 [1.8] years). Associations between ultraprocessed foods consumption and mortality are shown in Table 3. After adjusting for a range of covariates (model 3), we found that an increase in ultraprocessed foods consumption was positively associated with all-cause mortality; the HR per 10% increment of ultraprocessed foods consumed was 1.14 (95% CI, 1.04-1.27; P = .008). Similar results were observed with ultraprocessed foods proportion coded as quartiles and complete case analyses (eTables 7 and 8 in the Supplement). Associations between energy-weighted proportion of ultraprocessed foods and mortality are tabulated in eTable 9 in the Supplement.
These results were not substantially modified for the sensitivity analyses, carried out with further adjustment in model 3 for prevalent CVDs and cancer, with adjustment for a Western dietary pattern, and with the absolute amount (in weight) of ultraprocessed foods consumed as the exposure. When excluding death cases during the first 2 years of follow-up and when excluding prevalent CVDs and cancer cases, the associations were no longer statistically significant, probably owing to a loss of statistical power (data tabulated in eTables 10 and 11 in the Supplement).
Findings from this prospective study of a large French cohort suggest for the first time, to our knowledge, that an increased proportion of ultraprocessed foods in the diet is associated with a higher risk of overall mortality. Other prospective studies previously observed that ultraprocessed foods intake was associated with higher incidence of dyslipidemia7 and higher risk of obesity,8 hypertension,9 and cancer.10
In our sample, ultraprocessed foods accounted for 14.4% of the total weight of food and drink consumed and 29.1% of total energy intake. This proportion of energy intake derived from ultraprocessed foods was lower than in other industrialized countries,32-34 and it may be a consequence of participants’ dietary habits that involve high consumption of unprocessed foods.6,35 In line with previous findings, the proportion of ultraprocessed foods in the diet varied according to a person’s socioeconomic profile and lifestyle.32,36,37 These results underline the social inequalities associated with food choices. Ultraprocessed foods are attractive products for consumers as they are widely available in supermarkets,3,4 affordable, highly marketed, ready to eat, and made with a long shelf life.1
Several hypotheses could explain the associations between increasing ultraprocessed foods consumption and higher mortality risk. First, some of these foods have a high salt content, and high sodium intake has been associated with cardiovascular deaths38 and increased stomach cancer risk.12 Likewise, ultraprocessed foods tend to contribute to excessive intake of added sugar,33 whereas added sugar consumption has been associated with an increased risk for CVD mortality.39 Sugary beverages and processed meats may likely be factors in the observed correlation between ultraprocessed foods and mortality, as they have been consistently associated with mortality in prospective studies.40-43 Conversely, ultraprocessed foods have generally little fiber content, and higher ultraprocessed foods consumption was associated with lower fiber intake, whereas dietary fiber was previously linked with a substantially decreased mortality risk.44,45 Studies have documented that high ultraprocessed foods consumption was associated with unhealthy dietary patterns involving high intake of calories, fats, sugars, and salt.32,33,46-49 Those dietary factors could be associated with the development of noncommunicable diseases, which further lead to higher mortality risk.50,51 Studies have estimated that reducing saturated and trans fats, salt, and added sugar from the diet could present major advantages, such as preventing cardiovascular deaths.52,53
Associations between ultraprocessed foods consumption and mortality risk were somewhat weakened after adjustment for the modified PNNS-GS, suggesting that the overall nutritional quality of the diet may have a confounding role in the observed association. Even so, the results remained statistically significant after this adjustment; thus, we raise other hypotheses to explain these findings.
First, studies have documented the carcinogenicity of exposure to neoformed contaminants found in foods that have undergone high-temperature processing.54 The European Food Safety Authority stated in 2015 that acrylamide was suspected to be carcinogenic and genotoxic,13 and the International Agency for Research on Cancer classified acrylamide as “probably carcinogenic to humans” (group 2A).13(p11) Some studies reported a modest association between dietary acrylamide and renal or endometrial cancer risk.54 Further research is necessary to confirm these speculative hypotheses. Similarly, meat processing can produce carcinogens. The International Agency for Research on Cancer reported in 2015 that processed meat consumption was carcinogenic to humans (group 1), citing sufficient evidence for colorectal cancer.55 Moreover, the agency found a positive association between processed meat consumption and stomach cancer.55
In addition, ultraprocessed foods are characterized by the frequent use of additives in their formulations, and some studies have raised concerns about the health consequences of food additives. For instance, titanium dioxide is widely used by the food industry. However, findings from experimental studies suggest that daily intake of titanium dioxide may be associated with an increased risk of chronic intestinal inflammation and carcinogenesis.14,56,57 Likewise, experimental studies have suggested that consumption of emulsifiers could alter the composition of the gut microbiota, therefore promoting low-grade inflammation in the intestine and enhancing cancer induction58 and metabolic syndrome.15 In addition, some findings suggest that artificial intense sweeteners could alter microbiota and be linked with the onset of type 2 diabetes and metabolic diseases, which are major causes of premature mortality.59
Food packaging is also suspected to have endocrine-disrupting properties.60,61 During storage and transportation of food products, chemicals from food-contact articles can migrate into food, some of which might negatively affect health, such as bisphenol A.62 Epidemiologic data have suggested that endocrine disruptors are associated with an increased risk of endocrine cancers and metabolic diseases, such as diabetes and obesity.61
Concerns about the increased consumption of ultraprocessed foods have led to the September 2018 presentation of a parliamentary report to the French National Assembly. The report emphasized policy actions aimed at improving the nutritional quality of the food supply. In addition, several proposals pertaining to ultraprocessed foods were made, such as product reformulation (eg, limitation in bread salt content, restricted use of additives), reduction of neoformed contaminants and packaging residues found in industrially manufactured food products, and marketing restrictions on ultraprocessed foods.63 In accordance with a precautionary principle, the objective of reducing ultraprocessed foods consumption by 20% between 2018 and 2021 was introduced for the French 2018-2022 public health nutritional policy.64
Strengths of this study include its prospective design, the use of validated and repeated 24-hour dietary records accounting for intraindividual variability,25-27 and adjustment for the overall nutritional quality of the diet. Moreover, a large range of confounding factors were considered.
However, the study also has some limitations. First, the sample was taken from the NutriNet-Santé Study, which comprises a voluntary cohort of adults with long-term involvement in this nutrition study; that is, the participants are more health conscious than the general population.65 We selected participants who provided at least three 24-hour dietary records, which enabled us to increase the quality of dietary data collected, but we may have overrepresented compliant and healthy individuals. Thus, both the mortality rates and ultraprocessed foods intake amounts are probably lower than those in the general population, and consequently we may have underestimated the association observed. Overall caution is warranted when generalizing these results. Second, the risk of reverse causation is a general concern in observational mortality studies because individuals may change their diet after they develop a chronic disease. Exclusion of deaths during the first year has helped to minimize this bias. Reverse causation might be biased toward the null hypothesis (rather than inflate estimates), given that individuals with noncommunicable diseases may have modified their diet to follow nutritional recommendations (less consumption of ultraprocessed foods is likely).
Third, access to the type of food processing was not systematic for all of the food items. Each food product was categorized into the most likely NOVA group, but we cannot rule out some misclassification of food items at the individual level. This nondifferential classification bias may have led to an underestimation of the association observed between ultraprocessed foods consumption and mortality. The low number of death cases may have restricted our ability to detect some of the suspected associations. The statistical power was limited by the low number of CVD (n = 34) and cancer (n = 219) deaths, which did not allow us to perform the analysis for cause-specific mortality. The follow-up was rather short (median 7.1 years), although diet-related noncommunicable diseases develop over decades. Lastly, despite accounting for a wide range of confounding factors, we cannot exclude the hypothesis of residual confounding because of the observational design of this study, with its modest HRs and statistical significance. Thus, no causality can be established for the observed associations.
To tackle the issue of residual confounding, we performed models of incremental adjustment to observe the association independently from the nutritional characteristics of ultraprocessed foods and to limit the risk of overadjusted models. To test the robustness of the results in view of the risk for confounding factors resulting from the association between ultraprocessed foods and less healthy diets, we performed sensitivity analyses with adjustment for a Western dietary pattern. These results were consistent with the main analysis findings.
To our knowledge, this prospective study was the first to investigate the association between ultraprocessed foods consumption and mortality risk in a large population-based French cohort. Results suggested a positive association between increased ultraprocessed foods consumption and all-cause mortality risk. Further studies are needed to confirm those results in different populations and to disentangle the various mechanisms by which ultraprocessed foods may affect health, including both their nutritional features and their food processing–related characteristics. Ultraprocessed foods consumption has largely increased during the past several decades and may drive a growing burden of noncommunicable disease deaths.
Accepted for Publication: October 29, 2018.
Corresponding Author: Laure Schnabel, MD, MSc, Université Paris 13, Equipe de Recherche en Epidémiologie Nutritionnelle, Centre de Recherche Epidémiologie et Statistique Sorbonne Paris Cité, Inserm U1153, Inra U1125, Cnam, COMUE Sorbonne Paris Cité, UFR SMBH Paris 13, 74 rue Marcel Cachin, Bobigny F-93017, France (firstname.lastname@example.org).
Published Online: February 11, 2019. doi:10.1001/jamainternmed.2018.7289
Author Contributions: Dr Julia had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Dr Julia is a guarantor of the article.
Concept and design: Schnabel, Kesse-Guyot, Allès, Touvier, Buscail, Julia.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Schnabel, Kesse-Guyot.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Schnabel, Kesse-Guyot, Srour, Julia.
Obtained funding: Kesse-Guyot, Hercberg.
Administrative, technical, or material support: Kesse-Guyot, Touvier, Hercberg, Julia.
Supervision: Kesse-Guyot, Touvier, Hercberg, Buscail, Julia.
Conflict of Interest Disclosures: None reported.
Additional Contributions: We thank the scientists, dietitians, technicians, and assistants for their technical contributions to the NutriNet-Santé Study as well as all dedicated and conscientious volunteers. We are especially grateful to the following individuals from Université Paris 13, Equipe de Recherche en Epidémiologie Nutritionnelle, Centre de Recherche Epidémiologie et Statistique, Sorbonne Paris Cité: Nathalie Druesne-Pecollo, PhD; Younes Esseddik, MSc; Régis Gatibelza, MSc; Frédéric Coffinières, MSc; Paul Flanzy, MSc; Thi Hong Van Duong, MSc; Véronique Gourlet, PhD; Fabien Szabo, PhD; Nathalie Arnault, MSc; Julien Allègre, MSc; Laurent Bourhis, MSc; Cédric Agaësse; and Anne-Elise Dussoulier. The named individuals are paid members of the NutriNet-Santé Study research team; they received no specific incentives for their contribution to the current study.