Sorbonne Paris Cité, Epidemiology and Statistics Research Center, Nutritional Epidemiology Research Team (EREN), Inserm U1153; Inra U1125; Cnam; Paris 13, 7 and 5 Universities,

SMBH Paris 13, 74, rue Marcel Cachin, 93017 Bobigny Cedex, FranceTel: +33.148.388.968 ; Fax: +33.148.388.931 ; e-mail: b.srour@eren.smbh.univ-paris13.fr

Editor, Dr. Tiago Villanueva,The British Medical Journal

October 24th, 2018

Dear Dr. Villanueva,

Please find enclosed the revised version of our manuscript BMJ.2018.046135 titled ‘Ultra-processed food intake and cardiovascular disease risk in the NutriNet-Santé prospective cohort’.

We would like to sincerely thank you, the Editorial team, and all the reviewers for your valuable comments and for the opportunity that you gave us to revise this manuscript. All comments from the Editors and the Reviewers have been carefully taken into account. Changes have been highlighted throughout the text (red font). Also, please find enclosed the Editor and Reviewers’ comments followed by our point-by-point responses.

We have included additional sensitivity analyses as recommended; we have also detailed several points in the manuscript and have added one appendix (online supplemental data, 9 Appendixes in total) and one figure (the Flowchart of the study sample). Main results remain very consistent across all these further analyses, which strengthens the robustness of our findings.

We hope that our modifications adequately address the Editor and Reviewers’ comments and that our paper is now suitable for publication in The BMJ. We thank you again for your consideration.

Yours sincerely,

Dr Bernard Srour, PharmD(Corresponding author)

2

24-Sep-2018

Dear Dr. Srour

Manuscript ID BMJ.2018.046135 entitled "Ultra-processed food intake and cardiovascular disease risk in the NutriNet-Santé prospective cohort"

Thank you for sending us your paper. We sent it for external peer review and discussed it at our manuscript committee meeting. We recognise its potential importance and relevance to general medical readers, but I am afraid that we have not yet been able to reach a final decision on it because several important aspects of the work still need clarifying.

We hope very much that you will be willing and able to revise your paper as explained below in the report from the manuscript meeting, so that we will be in a better position to understand your study and decide whether the BMJ is the right journal for it. We are looking forward to reading the revised version and, we hope, reaching a decision.

Please remember that the author list and order were finalised upon initial submission, and reviewers and editors judged the paper in light of this information, particularly regarding any competing interests. If authors are later added to a paper this process is subverted. In that case, we reserve the right to rescind any previous decision or return the paper to the review process. Please also remember that we reserve the right to require formation of an authorship group when there are a large number of authors.

Thanks!

Tiago VillanuevaAssociate Editortvillanueva@bmj.com

*** PLEASE NOTE: This is a two-step process. After clicking on the link, you will be directed to a webpage to confirm. ***

https://mc.manuscriptcentral.com/bmj?URL_MASK=2dabc7658eca483ca195065ed9d7b8cd

**Report from The BMJ’s manuscript committee meeting**

These comments are an attempt to summarise the discussions at the manuscript meeting. They are not an exact transcript.

3

Members of the committee were: Sophie Cook (chair), Richard Riley (statistician), Wim Weber, Georg Roggla, Daoxin Yin, Tiago Villanueva

Decision: Put points

Detailed comments from the meeting:

First, please revise your paper to respond to all of the comments by the reviewers. Their reports are available at the end of this letter, below.--> We would like to sincerely thank the Editors and Reviewers for their valuable comments and suggestions. All of them have been carefully taken into account. We believe that they truly contribute to improve the quality, the overall presentation and the clarity of our manuscript. Below, you will find a point by point description of how each comment was addressed in the manuscript.

Please also respond to these additional comments by the committee:

- Our statistician made the following comments:Generally seems very good Their main analysis uses the missing indicator variable – but then they do imputation in a sensitivity analysis. The latter should just be the main analysis. That being said, the findings appear similar and so maybe this is a minor point, but at least the results should be in the main paper, not just supp material. --> Findings were indeed very similar (probably due to the very low proportion of missing values). As suggested by the Statistician Expert, multiple imputation results have now been moved to the results section of the main paper page 10.

I do not see that they have evaluated non-linear trends in the association, e.g. between ultra-processed food intake proportion and risk of death. --> The assumption of linearity between ultra-processed food consumption and cardiovascular disease risk was verified using restricted cubic spline (RCS) functions using the SAS® macro written by Desquilbet and Mariotti (1). This has now been added to the methods section of the manuscript on page 7.Please find below the corresponding plot. It has been added to the supplemental material, appendix 4.

4

I do not see results of checking the proportional hazards assumption.

--> Please find below the results of the graphical verification of the proportional hazards assumption (Log-log survival vs. log-time plots) mentioned in the Methods section. The brown curve (corresponding to the value 4) represents the fourth quartile of the distribution of the proportion of ultra-processed food in the diet (main exposure). Participants of this quartile have a higher risk of cardiovascular diseases during the whole follow-up period. This has been added to the supplemental material, appendix 4.

5

- Several editors were supportive since it looks novel and the study addresses an interesting research question.--> We sincerely thank the BMJ’s manuscript committee for this positive comment.

In your response please provide, point by point, your replies to the comments made by the reviewers and the editors, explaining how you have dealt with them in the paper.

Comments from Reviewers

Reviewer: 1

Recommendation:

Comments:This is a thorough premise and conclusion. The data shows the importance of basic, healthy foods to our diet and how the consumption of ultraprocessed foods is detrimental to all.Thank you for your work.--> We are grateful to Reviewer 1 for this positive comment.

6

Additional Questions:Please enter your name: Wendean C MarshJob Title: Operations Program AssociateInstitution: State of WisconsinReimbursement for attending a symposium?: NoA fee for speaking?: NoA fee for organising education?: NoFunds for research?: NoFunds for a member of staff?: NoFees for consulting?: NoHave you in the past five years been employed by an organisation that mayin any way gain or lose financially from the publication of this paper?: NoDo you hold any stocks or shares in an organisation that may in any waygain or lose financially from the publication of this paper?: NoIf you have any competing interests <A HREF='http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/declaration-competing-interests'target='_new'> (please see BMJ policy) </a>please declare them here: I do not have any competing interests.

Reviewer: 2

Recommendation:

Comments:This is an important article as it is the first one to explore the association between ultra-processed food consumption and cardiovascular disease, previous studies having focused on the association between ultra-processed food consumption and risk factors for cardiovascular disease such as dyslipidaemia (Rauber et al. 2015), hypertension (Mendonca et al. 2017), metabolic syndrome (Tavares et al. 2012; Lavigne-Robichaud et al. 2018), and overweight/obesity (Louzada et al. 2015; Mendonca et al. 2016; Juul et al. 2018) [among other health outcomes such as cancer (Fiolet et al. 2018)]. In addition, by using a prospective cohort design this study will better control reverse causality than previously carried out cross-sectional studies. This study adds knowledge to a pool of evidence from different countries showing that ultra-processed food consumption can be harmful for health (see previously mentioned references) and decrease diet quality (Monteiro et al. 2011; Moubarac et al. 2013; Poti et al. 2015; Louzada et al. 2015a; Louzada et al. 2015b; Moubarac et al. 2016; Luiten et al. 2016; Martinez Steele et al. 2016; Martínez Steele et al. 2017a; Martínez Steele et al. 2017b; Cediel et al. 2018; Julia et al. 2018; Louzada et al. 2018; Martínez Steele et al. 2018).The results from this study are important for the general population (we are all exposed and potential consumers of ultra-processed foods and at potential risk of CVD), for patients who already have CVD and policy makers who based on these results, may work on policies to diminish or control exposure to ultra-processed foods.

7

The research question is clearly defined and appropriately answered. The overall design of the study (prospective, cohort study) is appropriate to answer the research question (estimate the association between ultra-processed food consumption and incidence of CVD).Inclusion (participants aged over 18 years with access to the Internet) and exclusion criteria (less than 2 24h-dietary records during the first two years, with CVD at baseline, energy under-reporters) are clearly described (also in Appendix 2- Flowchart).The study is based on a convenience sample in which women (i.e. 80 women vs 20% men) and probably health-conscious individuals are well more represented, which could lead to selection bias and lack of generalizability. Still these limitations are somewhat compensated by a very big sample size (more than 100,000) and by adjusting for sex, education and other health behaviors.

--> We would like to thank reviewer 2 for this valuable and interesting comment. Indeed, as in most population-based etiological cohorts worldwide, participants included in NutriNet-Santé are volunteers recruited from the general population. As such they may be more concerned about their diet and health and adopt healthier nutritional habits than the general population (2). Although no nationally representative data is available for France regarding the proportion of ultra-processed foods in the diet, our study population may consume lower amounts of ultra-processed foods compared to their French counterparts. However, in an etiological study, more than representativeness, the most important feature is the broad representation of the diversity of behaviors and participants characteristics. We were able to grasp the high variability in the dietary intakes of the population, as shown by the large difference in ultra-processed foods consumption between the first and last quartiles of consumption. Even if caution is warranted for the extrapolation of our results to the general population (as pointed out in our discussion page 13), we assume that the observed association could even be stronger in the general population with more contrasted ultra-processed food intakes. We have added some discussion about this point page 13.

The Methods are clearly described. Cases were defined as first incidence of overall CVD, cerebrovascular diseases (stroke and transitory ischemic attack) and coronary heart diseases (myocardial infarction, angioplasty and acute coronary syndromes). Cases were defined according to International Chronic Diseases classification, based on self-report and information obtained through linkage with medical and mortality databases.The main exposure (gram contribution of ultra-processed foods per day, used as continuous or categorized according to sex-specific quartiles) and covariates (age, sex, BMI, physical activity, smoking status, number of 24-hour dietary records, alcohol, energy intake, family history of CVD and education; saturated fatty acids, sodium and sugar intakes; Healthy dietary pattern; baseline prevalent type II diabetes, dyslipidaemia, hypertension, and hypertriglyceridemia and treatment for these conditions) were clearly defined. Less than 5% data on covariates were missing and imputed. The proportion of IPAQ missing values was 14% and a category for missing values was created.The study was ethical and approved by IRB.The results are complete, well and clearly presented and answer the research question.Discussion and conclusions focus on the data and the message is clear. Key findings and limitations (including selection bias and generalizability, residual confounding and causality,

8

statistical power to detect some associations, length of follow-up and NOVA misclassification) are discussed. The findings are discussed in the light of previous evidence as well. References are up to date and relevant. Some suggestions have been included below.Both abstract and strobe accurately reflect what is stated in the paper.--> We thank reviewer 2 for these positive inputs.

Below I include some minor suggestions.- Page 5, line 16: May I suggest including an additional reference for “Worlwide, during the past decades, the consumption of ultra-processed foods has drastically increased (7-11)”: Reference 33 (Juul F, Hemmingsson E. Trends in consumption of ultra-processed foods and obesity in Sweden between 1960 and 2010. Public Health Nutr. 2015 Dec;18(1475–2727 (Electronic)):3096–107).--> The requested reference has been added.

- Page 5, line 18: I suggest including the following additional reference (56.8% of total calories come from ultra-processed foods in the UK) for “ultra-processed foods represent between 25% and 60% of total daily energy intake (12–22)”: Rauber F, Louzada MLC, Steele EM, Millett C, Monteiro CA, Levy RB. Ultra-processed food consumption and chronic non-communicable diseases-related dietary nutrient profile in the UK (2008-2014). Nutrients 2018; 10: 587.--> The requested reference has been added.

- Page 6, line 8: For clarity, may I suggest changing “participants aged over 18 years” for “participants aged 18 years or above”.--> We have made the modification page 5.

- Page 6, line 28: When you mention “To be included in the nutrition component of the NutriNet-Santé cohort, only two dietary records were mandatory”, do you refer to two dietary records overall or per series of 3 dietary records? Even though this becomes clearer further onwards in the text (two during the overall baseline period), I suggest clarifying this here.--> Indeed, two dietary records during the overall baseline period were mandatory for participants included in this study. We have clarified this point in the revised manuscript page 5.

- Page 6, line 31: Where there up to 12 records per participant? If so, I suggest this is specified in the text as follows: “intakes from all the 24h-dietary records available during the first two years of each participant’s follow-up (up to 12 records) were averaged”.--> There were up to 15 records per participants (up to 3 every six months from baseline until baseline+2 years). We have clarified this point in the revised manuscript page 5.

- Page 6, line 38: Please explain what “20%” refers to in the following statement “metabolic rate and Goldberg cut-off, and under-energy reporters (20%) were excluded (53).”.--> We apologize for the unclarity of this statement. The “20%” refers to the percentage of participants which were flagged as under-energy reporters and were therefore excluded from

9

the study. For more details, please refer to comment of reviewer 3 below. We have now clarified this point in the manuscript, methods section, page 5.

- Page 7, line 17: I think the following information now included in Appendix 1, is sufficiently important to be included in the manuscript “All food and beverage items of the NutriNet-Santé composition table were categorized by a team of three trained dieticians into one of the four food groups in NOVA, a food classification system based on the extent and purpose of industrial food processing (1–3). The whole classification was then reviewed by a committee composed of the three dietitians and five researchers, specialists in nutritional epidemiology. In case of uncertainty for a given food/beverage item, a consensus was reached among researchers based on the percentage of home-made and artisanal foods versus industrial brands reported by the participants.”--> We have now included this information in the manuscript on pages 5-6.

- Page 7, line 54: Why were sex-specific quartiles of the proportion of ultra-processed food rather than age-specific quartiles (or just total quartiles) used? I suggest explaining this in the manuscript. --> As usually done in nutritional epidemiology, we have chosen to use sex-specific quartiles of the proportion of ultra-processed foods since women generally tend to have a healthier diet than men. Furthermore, this choice ensures we have the same sex-ratio between our four quartiles. This has been clarified in the manuscript page 7.The Cox models used in this study are based on age as a time scale, which ensures proper adjustment for age. It was not necessary to build age-specific quartiles. Besides older participants tended to have healthier diets than younger ones, and in particular, lower ultra-processed food consumption (table 1). Creating age-specific quartiles would therefore have led to very heterogeneous quartiles with high intra-quartile differences in ultra-processed food consumption. This could have led to lower contrasts between quartiles, and eventually an artificial underestimation of the associations observed.

- Page 8, line 2: In the following statement “and incidence of overall CVD, cerebrovascular diseases (stroke and TIA) and coronary heart diseases (MI and angioplasty)” shouldn´t “acute coronary syndromes” be mentioned as “coronary heart diseases”?--> We apologize for this omission. This has been modified in the manuscript page 7.

- Page 9, line 1: I would suggest “less highly educated” (or “with less higher education”) instead of “less educated” because according to Table 1, both the proportion of less educated and more highly educated seem to decrease with quartiles of ultra-processed food consumption.--> We have clarified this point page 8.

- Page 9, line 3: In accordance with Table 1, I suggest adding “along with less fruit and vegetable consumption and less dietary fiber intake” to the statement “Furthermore, they had higher body mass index, intakes of energy, lipids, carbohydrates and sodium, along with lower alcohol intake and prevalence of metabolic diseases”.

10

--> We have added this statement as requested page 8.

- Page 9, line 17: In regards to statement “During follow-up (518208 person-years, median follow-up time=5.2y), 1409 first incident CVD events occurred, among which 106 MI, 485 angioplasties, 73 ACS, 155 strokes and 674 TIA.”, shouldn´t these three numbers “106 MI, 485 angioplasties, 73 ACS” add up to 665 (and not 664), in accordance with Table 2?

--> We apologize for this typo. The number of ACS cases is 74 not 73. This has been corrected in the manuscript, page 9.

- Page 9, line 47: Why weren´t the results of models after restriction of the population study to the participants with at least six 24h dietary records during the first two years of follow-up described in Results (and/or Appendix Table 8)?--> We apologize for this omission. The results are now added in the manuscript page 9.

- Page 10, line 22: An important word is missing in the following phrase “In addition, several food groups that are mainly ultra-processed and are largely XXX have been associated with increased risks of cardiometabolic outcomes with a high concordance, i.e. sugar-sweetened beverages and processed meats (2).”--> We apologize for this typo. The original sentence was supposed to be “In addition, several food groups that are mainly ultra-processed and are largely consumed in Western-type diets have been associated with increased risks of cardiometabolic outcomes with a high concordance”. This has been corrected in the manuscript page 11.

- Page 11, line 29: I think the following limitation could be further discussed: “Second, some misclassification in the NOVA ‘ultra-processed food’ category cannot be ruled out”. Would misclassification lead to an under- or an overestimation of ultra-processed food consumption? How will this misclassification affect the studied association?--> Indeed, misclassification in the NOVA ultra-processed category can never be ruled out. However, the committee who performed/reviewed the classification (composed of the three dietitians and five researchers) tried to avoid any unidirectional and systematic bias. Any remaining classification mistake could have led to a non-differential measurement error (identically in future cases and non-cases), most probably leading to an under-estimation of the observed associations. Some discussion about this point has been added to the manuscript page 12.

- Page 11, line 37: I am not sure whether I understood the link between the following two statements: “Fourth, the length of follow-up was relatively limited in time, since the cohort was launched in 2009” and “Still, a classic assumption in nutritional epidemiology is that the measured exposure at baseline (especially since we averaged a two-year period of exposure) actually reflects more generally the usual eating habits of the individual not only at the moment of the study but also several years prior to their inclusion in the cohort”.--> Since the measured exposure at baseline in middle-aged adults generally reflects more broadly the usual eating habits of the individuals (not only at the moment of the study but also

11

several years prior to their inclusion in the cohort), we assume that our study provided insights into the associations between “chronic” and mid-to-long term consumption of ultra-processed foods and cardiovascular risk. We have clarified this point in the Discussion page 13.

- Page 11, line 43: Regarding statement “Last, as it is usually the case in volunteer-based cohorts, participants to the NutriNet-Santé cohort were more often women, with health-conscious behaviors and higher socio-professional and educational levels as compared to the general French population (66). This might limit the generalizability of the findings and probably lead to a likely underestimation of the strength of the associations, even though an overestimation cannot be excluded”, how would selection bias lead to an underestimation of the strength of the association between ultra-processed food consumption and incidence of CVD? Is it because high ultra-processed food consumers are underrepresented? Or because losses to follow-up vary according to ultra-processed food consumption, being higher among the higher quartiles? --> Due to the characteristics of the participants (more health-conscious, better educated, etc. than the overall French population), ultra-processed food consumers may have been underrepresented in this study. This might have led to an underestimation of the strength of the association between ultra-processed food consumption and incidence of CVD, by reducing the contrast between the first and the last quartiles of ultra-processed food consumption. We have clarified this statement in the manuscript page 13.

- Page 11, line 52: To make sense of this phrase “According to the national INCA3 study conducted by the French Food safety Agency in 2016 (7), only about 1/3 of transformed foods (e.g. ice cream, biscuits, sweet pastries, dairy desserts) was homemade”, and link it with the previous phrase “To date, no nationally representative data is available regarding the proportion of ultra-processed food in the diet in the French population, thus comparison with our population study is not straightforward” I think you would need to provide a definition (and not just examples) of “transformed foods” according to the French Food Safety Agency.--> In the nationally representative INCA3 study, no clear definition is provided by the authors for “transformed” foods. It was not based on the NOVA classification. They provide a list of all food groups that they considered as “transformed”: sweet pastries, biscuits, dairy desserts, ice cream, fruit purée and fruit in syrup, fruit and vegetable juices, soups and broths, sandwiches, pizzas and salted pastries, as well as mixed dishes composed of egg, meat, fish, vegetable and/or starchy foods (cereals, legumes or potatoes). More than half of the “transformed” foods consumed outside catering establishments by adults aged 18-79 were manufactured industrially (about one-third were homemade, while the rest was handcrafted, e.g. caterer). We have clarified this point in the discussion pages 13-14.

- Page 20, line 50: Should you not replace in Table 1 footnote c “by x² test” for “by ANOVA or χ2 tests”?--> We apologize. We have now replaced x² test by ANOVA or χ2 tests in Table 1 footnote c.

Additional Questions:Please enter your name: Eurídice Martínez Steele

12

Job Title: Posdoctorate fellowInstitution: Faculdade de Saude Publica, Universidade de Sao PauloReimbursement for attending a symposium?: NoA fee for speaking?: NoA fee for organising education?: NoFunds for research?: NoFunds for a member of staff?: NoFees for consulting?: NoHave you in the past five years been employed by an organisation that mayin any way gain or lose financially from the publication of this paper?: NoDo you hold any stocks or shares in an organisation that may in any waygain or lose financially from the publication of this paper?: NoIf you have any competing interests <A HREF='http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/declaration-competing-interests'target='_new'> (please see BMJ policy) </a>please declare them here:

Reviewer: 3

Recommendation:

Comments:Dear Editor

The manuscript is relevant, original, brings a novel evidence on this important field.It is very well written, it has very high-quality analyses and robust results. I strongly recommend the publication.--> We sincerely thank reviewer 3 for this positive comment.

Some minor suggestions:Page 3 and abstract - I don´t believe that the “precautionary principle” applies here. It may be used when you have a hypothesis not empirically tested. It is not the case. I suggest you keep the sentence as “Considering other studies that have shown associations between ultra-processed food consumption and other non-communicable diseases, the consumption of fresh or minimally processed foods should be promoted”. The same applies for the abstract.--> We have now removed the “precautionary principle” from the box and the abstract.

Introduction

Page 4, line 10. I suggest you drop the reference number 4 on high-risk patients and keep just the references from WHO and The European and American Guidelines (references number 1, 5 and 6).--> Previous reference 4 has been dropped.

13

Page 4, line 15 – Reference 7 is not fitting well here since it does not provide evidence on trends of food consumption. Consider also citing “PAHO. Ultra-processed food and drink products in Latin America: Trends, impact on obesity, policy implications. Washington D.C.: PAHO, 2015.” and “Juul F, Hemmingsson E. Trends in consumption of ultra-processed foods and obesity in Sweden between 1960 and 2010. Public Health Nutrition 2015; 25:1-12.”--> Both suggestions have been taken into account on page 4.

Page 4, line 18 – Consider citing “Rauber, F. et al. Ultra-Processed Food Consumption and Chronic Non-Communicable Diseases-Related Dietary Nutrient Profile in the UK (2008-2014). Nutrients, v. 10, p. 587, 2018.”--> This reference has been added, page 4.

Page 4, line 23. You may want to add evidence on the higher energy density from the ultra-processed foods, the potential impact on satiety control and glycemic responses.--> We have added the notions of energy density, potential impact on satiety control and glycemic responses in the introduction page 4.

Page 4, line 23 “…highly palatable(12,23).” Reference 12 is not fitting well here.--> Reference 12 has been removed.

Methods

Please clarify how you have dealt with outliers. Did you drop out of the analyses 20% of the initial sample? (page 5, line 39).--> Energy underreporting was identified using Black’s method (3,4) based on the original method developed by Goldberg et al (5), relying on the hypothesis that energy expenditure and intake, when weight is stable, are equal. Black’s equations are based on an estimate of the person’s basal metabolic rate (BMR) calculated via Schofield’s equations (6) and taking into account sex, age, height and weight, as well as physical activity level (PAL), number of 24h records, intra-individual variabilities of reported energy intake and BMR, and intra/inter-variabilities of PAL. In the present study, intra-individual coefficients of variations for BMR and PAL were fixed using the values proposed by Black et al., i.e. 8.5 % and 15%, respectively. For identifying under-reporters, the 1.55 value of PAL was used. It corresponds to the WHO value for “light” activity, which is the probable minimum energy requirement for a normally active but sedentary individual (not sick, disabled or frail elderly). A higher value might have exaggerated the extent of under-reporting. Some under-reporting individuals were not excluded if their reported energy intake, initially estimated abnormally low, was found to be likely in case of recent weight variation or reported practice of weight-loss restrictive diet or proactive statement of the participant that he/she ate less than usual on the day of the dietary record. In the cohort, 20.0 % of the subjects were considered as under-reporters and were excluded from the study. This detailed presentation of the method for detection of under-reporters is now presented in appendix 1. Besides, we have clarified what the “20.0%” referred to in the manuscript page 5.

14

Please provide more information on sodium intake assessment (page 5, line 43)--> Sodium intake was assessed via a specific module included in the 24 hour records, taking into account native sodium in foods, salt added during the cooking, and salt added on the plate. It has been validated against sodium urinary excretion biomarkers (7). We have added these clarifications in the manuscript page 5.

Suggest changing “Degree of food processing” to “Extent and purpose of food processing”--> This suggestion has been taken into account on page 5

Statistical analysis

BMI is an important mediator of the association between ultra-processed foods intake and CVD. It would be important to present a model without this adjustment (not only as sensitivity analyses). A suggestion would be to include this variable on model 4. --> We have now added a model (model 4) without adjustment for BMI in Table 2.

Results

Page 8, line 7 Please better explain and/or give examples of “sugary products” (Cookies? Industrialized desserts?), “ultra-processed fruits and vegetables” (?), “starchy foods” (?), and “processed meats and fish” (nuggets? sausages?)--> Some examples have been added to the results page 9: “Main food groups contributing to ultra-processed food intake were sugary products (28%, e.g. confectionaries, ice-cream, pastries, sweetened dairy desserts) followed by ultra-processed fruits and vegetables (18%, e.g. instant powder dehydrated vegetable soups and broths, vegetable nuggets, fruit-based sweetened desserts), beverages (16%, e.g. sodas, sugary and artificially sweetened non-carbonated beverages), starchy foods and breakfast cereals (12%, e.g., pre-packaged bread, industrial dough, ready-to-eat industrial pasta or potato plates, breakfast cereals), and processed meat and fish (11%, e.g., nuggets, fish fingers, sausages, processed ham) (Figure 2).”

Page 8, line 11 – You did not describe the Food Standard Agency Nutrient Profiling in your methods section. Actually, it is not clear why you did this analysis. I could drop this from the paper or better explain it in the text. --> In order to provide some information on the nutritional quality of ultra-processed foods, we have calculated their proportion across the different categories of the Nutri-score. This score, calculated based on a modified version of the Food Standard Agency Nutrient Profiling system has been endorsed by the French and Belgian Ministries of Health as the official nutrient profiling system in these countries. Details about its calculation are now provided in Appendix 3. We have added this information in the Methods section page 7 and reformulated the corresponding sentence in the results section page 9.

I suggest you do not present results of subgroups of ultra-processed foods (Appendix 7).

15

--> We understand the concern of the reviewer, as the effects of ultra-processed foods on human health may go through complex mechanisms involving synergic effects of a chronic exposure to multiple factors, including cocktails of food additives, neoformed compounds and contact materials. These mechanisms can hardly be dispatched based on food groups as they should be considered globally. However, we found it interesting to explore whether one ultra-processed food group entirely drove the whole association (which was not the case). Several other reviewers were also interested by this approach and even asked further details regarding food-group analyses. Thus, we were compelled to let the corresponding supplementary table in the appendix, but we have added some discussion about this point pages 12-13.

Discussion

You could add to your discussion of possible mechanisms that explain the association, the specific effects of ultra-processed beverages (sugar-sweetened beverages). They “damage” satiety control mechanisms that can lead to weight gain and there are also some evidence on possible damages advanced glycation end products present in soda.See: DIMEGLIO, D. P.; MATTES, R. D. Liquid versus solid carbohydrate: effects on food intake and body weight. Int J Obes Relat Metab Disord, v. 24, n. 6, p. 794-800, 2000. URIBARRI, J. et al. Single oral challenge by advanced glycation end products acutely impairs endothelial function in diabetic and nondiabetic subjects. Diabetes Care, v. 30, n. 10, p. 2579-82, 2007.--> We thank reviewer 3 for theses inputs: Indeed, these mechanisms are plausible biologically and might underlie the associations between ultra-processed foods and beverages and increased cardiovascular disease risk. Sugar-sweetened beverages might delay the trigger of the internal satiety signal, leading therefore to an excessive caloric ingestion (8). On the other hand, several ultra-processed foods and beverages (i.e. confectionery snacks, sugar-sweetened beverages, cakes, sports drinks, breakfast cereals) may contain relatively high levels of glucose-derived advanced glycation end-products (Glu-AGE) (9), which could over time lead to and/or accelerate vascular disease (10).We have added these points to the Discussion section on page 11.

Additional Questions:Please enter your name: Maria Laura da Costa LouzadaJob Title: ProfessorInstitution: UNIFESPReimbursement for attending a symposium?: NoA fee for speaking?: NoA fee for organising education?: NoFunds for research?: YesFunds for a member of staff?: Fees for consulting?: NoHave you in the past five years been employed by an organisation that may

16

in any way gain or lose financially from the publication of this paper?: NoDo you hold any stocks or shares in an organisation that may in any waygain or lose financially from the publication of this paper?: NoIf you have any competing interests <A HREF='http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/declaration-competing-interests'target='_new'> (please see BMJ policy) </a>please declare them here: No

Reviewer: 4

Recommendation:

Comments:Originality of the workThis study investigates the association between ultra-processed food consumption and cardiovascular disease in a large cohort in France. Previous studies on ultra-processed food have simply examined the consumption of ultra-processed foods in sub-groups of the population (e.g. Correa EN et al. Sao Paulo Med J 2018;200-207), or the association between consumption of ultra-processed food and other risk factors, such as BMI (e.g Juul F et al. Br J Nutr 2018;120:90-100 and Cunha DB et al. Nutr Diabetes 2018;8:28). There is hardly any literature on ultra-processed foods and disease outcomes. This research group has previously published on ultra-processed food and cancer risk in the same cohort and in the current paper they report on cardiovascular disease risk. As far as I can see, this is the first study to investigate the association between ultra-processed foods and risk of cardiovascular disease. This paper takes a different approach to traditional nutritional epidemiology papers, which normally look at the association between a macronutrient or micronutrient or a food group and disease risk. In the current paper, the exposure is the proportion of the diet that is ultra-processed. It is an interesting approach, and one that could perhaps be easier to translate into advice (i.e. limiting intake of ultra-processed food items, compared to say advice on intake of nutrients). On the other hand, I wonder what component of the ultra-processed food is responsible for these associations (is it the nutrient composition, or the other components?). Overall, I think this is an interesting way to look at the food environment and complements more traditional nutritional epidemiological studies.--> We would like to sincerely thank reviewer 4 for this positive comment. The associations between ultra-processed food and cardiovascular disease risk are complex and the actual state of the art does not allow to draw firm conclusions regarding the different possible mechanisms that underlie these associations: as we cannot deny that the nutritional quality of these products is lower than other foods and beverages, other studies as well as our statistical analyses suggest that other “non-nutritional” mechanisms are involved in these relationships. As discussed pages 11,12 and 13, they may involve cocktails of food additives, contact materials or neoformed compounds. The further exploration of the different mechanisms involved in the association between ultra-processed food and cardiovascular disease risk is a key perspective to the present study. In the meanwhile, we wish our study will add more significant evidence and knowledge to this new research field, and we hope that, thanks to the scientific community, we

17

researchers will be able in the few coming years to elucidate these associations and their underlying mechanisms.

Importance of work to general readers I think a general journal is a good fit for this work. ‘Ultra-processed foods’ is becoming an increasingly used term; however, there is a lack of research on the associations between ultra-processed foods and disease outcomes. I think this work would be of interest to clinicians and patients and to policymakers. --> We thank reviewer 4 for recommending the publication of this paper.

MethodsThe research question is clearly defined. The STROBE guidelines appear to have been followed.The participants are described briefly in this paper and the reader is referred to a previous paper which describes in the cohort in detail. The dietary assessment methods are clearly described. The authors included participants who had completed at least two 24-hour dietary records, I think this is an acceptable and pragmatic balance between ensuring sufficient number of participants in the study and trying to get an estimate of habitual or usual diet. Most participants had more than 2 records. The self-administered dietary assessment method has been validated. The classification of foods into ultra-processed and the other categories was done by a team of dieticians and nutritional epidemiologists and appears sensible. The way the exposure variable is categorised is by proportion of the diet on a weight basis (e.g. % of g/day). The authors argue this is a better choice than categorising based on energy because for example artificially sweetened beverages do not contain energy, but still contain additives that would classify them as processed – I agree with the authors approach. The study has ethical approval and I have no ethical issues with the study.

1. The flow chart is given in supplementary material but I would prefer it to be moved to the main paper.--> Flowchart has been moved from supplementary material to figure 1 in the main manuscript.

Outcome identificationAs I understand it, the participants self-report major health events through questionnaires. If an event is declared participants send medical records for review by study physicians. Also the cohort is linked to national health insurance administrative database and the national mortality registry, which should identify participants who may not have self-reported events. It may be possible that some events are missed although this dilution effect would not have a major effect on the results. If study physicians review medical records for those that self-report events this should minimise false positive reports (which is more of a concern in terms of impacts on the results).2. What proportion of the cohort can be linked to the insurance database?--> Participants were regularly asked to declare any new health event through the yearly health status questionnaire, through a specific check-up questionnaire (every three months) or at any time through a specific interface on the study website. For each incident major health event declared (CVD or cancer), participants were contacted by a physician of the team and asked to

18

provide any relevant medical records. Whenever necessary, the study physicians contacted the physician of the patient and/or hospitals to collect additional information. Afterwards, these medical data were reviewed by a physician expert committee. An investigation was also conducted by the physicians of the NutriNet-Santé study by contacting the participant’s family and/or his/her physician in case of no connection to the study website for more than one year. This system constitutes the main source of case ascertainment in the cohort.

When the NutriNet-Santé cohort was launched in 2009, the French legislation did not allow access to individual health insurance data for merging purposes with cohort data. We have gone through several administrative steps and procedures to change things for us and for the rest of the scientific community in France. As a consequence, our research team was the first to obtain the authorization by Decree in the Council of State (n°2013-175) to link data from our general population-based cohorts to medico-administrative databases of the National health insurance (SNIIRAM database). In 2013, participants were asked in an ad-hoc questionnaire whether they wish to provide their social security number or not. Medical data regarding 50,240 participants of the NutriNet-Santé study were linked to the insurance database. Thus, for all these individuals, medical data were completed by the additional information from these databases, thereby limiting potential bias due to people with CVD who may not report their disease to the study investigators. Since June 2013, new participants included in the cohort are automatically asked to provide their social security number in the baseline questionnaire or can provide it later on during follow-up, which increased acceptance rate of the procedure to about 60% (versus about 20% with pending authorization and 20% participants who declined the linkage). This system is very recent and corresponds to deep changing in the access to medico-administrative data in France for research purposes, to which our team contributed as a pioneer.

Importantly, we have verified that results remained similar when analyses included only cases and censored participants with linked SNIIRAM data (HRfor a 10-point increment in the proportion of ultra-

processed foods in the diet=1.13 (1.06-1.12) p=0.0003 for CVD risk).

Besides, although SNIIRAM databases are conceptually interesting, they allowed us to detect only 9 missing cardiovascular events, showing that our follow-up procedure in the NutriNet-Santé data is already performant. The SNIIRAM database theoretically covers all reimbursements of healthcare and hospitalizations by the social security. However, these data were not initially designed for research purposes and errors may be found in these databases: we have detected some cases reported in our questionnaires and validated on the basis of medical documents by our physicians, but which were not in the SNIIRAM databases (n = 23). The combination of these different collection methods for case ascertainment therefore maximizes case coverage, although exhaustiveness can never be guaranteed. Regarding deceased participants, medical information is collected using the SNIIRAM databases if a prior authorization has been given by the participant. Otherwise, the family and/or treating physician are contacted, and eventually, missing medical information is completed with an additional and exhaustive linkage to the French National cause-specific mortality registry

19

(CépiDC, which includes both dates and causes of death, and is accessible for all French citizens, without specific authorization or identification number).

Information about these points has been added to the manuscript in the Methods, Results, and Discussion sections pages 6, 10 and 13.

3. What are the ICD-10 codes that correspond to each of the outcomes (CVD, stroke, TIA, MI, and angioplasty)? --> The corresponding ICD-10 codes are the following:

- Stroke: I64- MI: I21- TIA: G45.8 and G45.9- ACS: I20.0 and I21.4- Angioplasty: Z95.8

We have added these codes to the Methods section on pages 6-7.

Results The tables are well presented and clear.--> Thank you for this positive comment.

Interpretation and conclusions My main concern about the findings would be around residual confounding by other factors. The authors do mention this in the limitations briefly, but perhaps this discussion could be expanded a bit. I found it interesting in Appendix 7, when looking at the specific categories of ultra-processed foods, that fats and sauces and salty snacks particularly have strong associations with CVD. It made me think that nutrition composition may be driving the associations with CVD, although the results were attenuated, but not by much when additional dietary factors were added to the model. --> This is an observational study, thus, residual confounding can never be entirely excluded. To limit this bias, a large number of potential confounders have been taken into account in the models. In addition, several sensitivity analyses (testing further adjustments and/or stratifications) are presented in the results and in the supplementary material, showing the stability of the results. Residual confounding may be related to unmeasured confounders, but also to imprecision in the measure of included confounders. For instance, in model 5, treatments for each metabolic disorder were considered as binary variables, since the duration of the treatment and the compliance were not measured. Same goes for prevalent metabolic disorders in model 5, duration since diagnosis was not measured. A paragraph acknowledging and discussing the possible residual confounding has been added in the limits section of the revised manuscript page 12.Several of the ultra-processed food groups were associated with CVD risk in Appendix 7 (now Appendix 8 in the revised version) (beverages, fats&sauces, meat, fish and egg, salty snacks, sugary products). Fats&sauces, and salty snacks, were indeed statistically significantly associated with CVD risk but the absolute value of the effect size should be considered with

20

caution since these food groups were relatively low contributors to ultraprocessed food intakes (5% and 2% respectively, Figure 2), thus confidence intervals associated with their HR was broad. Among the different hypotheses that could explain the observed association, the lower overall nutritional quality of ultra-processed foods can be advanced, but would not be the only one involved. Further adjustments for 1) several nutrients that potentially exert deleterious cardio-metabolic effects (saturated fatty acids, sodium and sugar), 2) dietary fiber, fruit and vegetable intakes, and 3) Western and Healthy dietary patterns slightly modified the associations, but they always remained strongly significant. The nutrition composition is therefore not the only driver of the associations. The presence of other compounds (food additives, substances formed during processing, materials in contact with food, etc.) could also play a role, as hypothesized in the discussion pages 11 to 13.

4. What is the basis for the statement that repeated 24-hour dietary records are more accurate than FFQs? Page 10, line 20? Do the authors have a reference to support this statement?--> In nutritional epidemiology in general, repeated dietary records are recognized to be more efficient than FFQs at placing individuals in the distribution of habitual diets, due partly to inaccuracies in the estimation of frequency of food consumptions in FFQs (11) and showed better correlations with urinary biomarkers (12). We have added a reference to support this statement on page 12, as requested.More specifically, FFQs provide interesting insights into usual nutritional exposure but are less appropriate in this particular study to properly discriminate foods according to their degree of processing (indeed, FFQs are often based on 100-200 food groups while >3300 food items were distinguished in the 24h dietary records used in NutriNet-Santé).

ReferencesReferences are appropriate, no obvious omissions.

Other minor comments:

5. Introduction, line 26, is it supposed to be ‘plant/animal food cells’?--> Indeed, we apologize for this typo. It has been corrected.

6. Discussion, page 9, line 31, were the results adjusted for Western-type dietary pattern, or only the healthy dietary pattern? The footnotes for table 2 suggest just the healthy dietary pattern.--> Both were tested. Results were adjusted for a Healthy-type dietary pattern (table 2) and for a Western-type dietary pattern (sensitivity analyses in Appendix 9).

Dr Kathryn BradburyAdditional Questions:Please enter your name: Kathryn BradburyJob Title: Senior Research Fellow

21

Institution: University of AucklandReimbursement for attending a symposium?: NoA fee for speaking?: NoA fee for organising education?: NoFunds for research?: NoFunds for a member of staff?: NoFees for consulting?: NoHave you in the past five years been employed by an organisation that mayin any way gain or lose financially from the publication of this paper?: NoDo you hold any stocks or shares in an organisation that may in any waygain or lose financially from the publication of this paper?: NoIf you have any competing interests <A HREF='http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/declaration-competing-interests'target='_new'> (please see BMJ policy) </a>please declare them here:

Reviewer: 5

Recommendation:

Comments:This study examined the association between ultra-processed food intake and the risk of cardiovascular diseases in a web-based sample. The study found that higher proportion of ultra-processed food intake was associated with higher risk of CVDs and the sub-types (CHD and cerebrovascular disease). The findings provided some useful evidence for the prevention of CVDs. However, some problems are needed to be addressed in this study. --> We thank Reviewer 5 for this overall positive comment. All comments and questions have been addressed below.

Major comments1. According to the objective of this study, the main hypothesis would be that the ultra-processes for foods would increase the risk of CVDs due to a poorer nutritional quality, a wide range of additives, and neoformed contaminants. To address this hypothesis, it is needed to compare the CVD risk of ultra-processed foods to those of “unprocessed or minimally processed foods” of the same types (e.g., meat). The control group (non-exposure) should be unprocessed or minimally processed foods in the same type, but not unprocessed or minimally processed other foods. The proportions of ultra-processed foods for each individual type of foods should be calculated. However, the authors compared the CVD risk by different proportion of ultra-processed foods in total foods. Such comparisons could not get the independent association of the “ultra-process” itself (A) from the ultra-processed (A) foods (B) due to different “B”. Therefore, the unfavorable associations between ultra-processed foods and the CVD risk might be (at least in part) explained by the specific types of foods rather than the ultra-processes. In this case, the main hypothesis was not well-addressed, and the conclusions might mislead the readers.

22

--> We understand this relevant and interesting question and thank the Reviewer 5 for asking it. Our answer will discuss two aspects.First, it is not possible to build a model where the high consumers of a given food group in its ultra-processed form would be compared to the high consumers of that same food group in its non-ultra-processed form, because these two behaviors are not exclusive. A same individual can consume important quantities of both ultra-processed and unprocessed food of the same group. Thus, this strategy does not provide a suitable control group, as individuals might be categorized into consumers and controls at the same time. For instance, in our study sample, >6300 individuals are high consumers (quartiles 4) of both ultra-processed meat, fish and eggs AND unprocessed meat, fish, and eggs. Besides, many ultra-processed foods do not really have a direct matching unprocessed equivalent (e.g. sodas, confectionery, energy bars). However, in order to comply with Reviewer’s 5 request, we have further developed the analyses presented in Appendix 8 by food groups. Previously, we only presented the associations between the consumption of food groups in their ultra-processed form and CVD risk and showed some increased risks for several ultra-processed food groups (beverages, fats and sauces, meats, sugary products and salty snacks). We have now added (to Appendix 8 and in the Results section page 9) the same analysis but for the corresponding food groups in their non-ultra-processed form. Interestingly and in line with our hypotheses, the associations observed for ultra-processed beverages, meat, fish and egg, sauces, and sugary products became non-significant for the corresponding non-ultra-processed foods groups. The association remained significant only for salty snacks, but confidence intervals were very broad due to a limited consumption of these foods in our population study. We have also added the proportions of ultra-processed foods for each food group in Appendix 8, as requested.The second aspect is a more general consideration. The effects of ultra-processed foods on human health may go through complex mechanisms involving synergic effects of many compounds and characteristics of ultra-processed foods. A chronic exposure to multiple factors, including cocktails of food additives (e.g. glutamate salts frequently used in sauces, artificial sweeteners frequently used in beverages, preservatives frequently used in ready-to-eat meals, etc…), neoformed compounds and contact materials may play a role in the studied association. These mechanisms can hardly be dispatched based on food groups as they should be considered globally. Creating an indicator for the proportion of ultra-processed foods in the diet allows distinguishing individuals with a high/low exposure to these cocktail interactions. The fact that the associations were stronger when considering the overall ultra-processed food proportion in the diet, rather than the associations in specific food groups argue in favor of these potential cocktail effects. A key perspective of the present study will be to go further in the understanding of underlying mechanisms and potentially deleterious factors. In particular, our research team is currently launching a large-scale epidemiological and experimental programme to investigate chronic exposure to food additives (single substances and multi-exposure cocktails) and its potential impacts on health (13). We have added some discussion about this point page 13.

2. The total amount of ultra-processed foods was calculated by adding the weight in eating form for each food directly. However, the densities of different types of ultra-processed foods are quite different (e.g., salty snacks vs. beverages). Although there is no ideal method to calculate the total amount of ultra-processed foods, sensitivity analyses are needed by using

23

different methods to pool the total amount of ultra-processed foods, and discuss the strength and limitations of the method selected for the conclusion.--> We have chosen to use weight and not energy to calculate the aggregated mean because (as also acknowledged by Reviewer 4), it was the best way to take into account ultra-processed foods with no energy content (i.e diet sodas).However, in order to comply with Reviewer 5’s suggestion, we have performed a sensitivity analysis by weighting the ultra-processed variable by the energy (% Kcal/day). The results remained stable: HRfor a 10-point increment in the proportion of ultra-processed foods in the diet weighted by energy=1.06 (1.01 to 1.12), P=0.01, for overall CVD risk. We have added this analysis to the Methods page 7 and Results section page 10 and have discussed the strengths/limitations of our method in the Discussion page 13.

3. The case ascertainment was based on participants’ reports and the medico-administrative databases of the SNIIRAM. The validity of these methods depends on the compliance and ability of the participants, and to what extent the participants’ data can be found in the databases. To address the first problem, the authors might compare the incidence of CVDs by different compliance groups (by days of dietary records). The second problem depends on the proportion of subjects provided their ID number, the coverage of the SNIIRAM databases, and the proportion of emigrants to other countries not covered by the SNIIRAM databases. The validity of the method for case ascertainment needs to discuss. --> As stated in our answer to point 2, Reviewer 4:“Participants were regularly asked to declare any new health event through the yearly health status questionnaire, through a specific check-up questionnaire (every three months) or at any time through a specific interface on the study website. For each incident major health event declared (CVD or cancer), participants were contacted by a physician of the team and asked to provide any relevant medical records. Whenever necessary, the study physicians contacted the physician of the patient and/or hospitals to collect additional information. Afterwards, these medical data were reviewed by a physician expert committee. An investigation was also conducted by the physicians of the NutriNet-Santé study by contacting the participant’s family and/or his/her physician in case of no connection to the study website for more than one year. This system constitutes the main source of case ascertainment in the cohort.

When the NutriNet-Santé cohort was launched in 2009, the French legislation did not allow access to individual health insurance data for merging purposes with cohort data. We have gone through several administrative steps and procedures to change things for us and for the rest of the scientific community in France. As a consequence, our research team was the first to obtain the authorization by Decree in the Council of State (n°2013-175) to link data from our general population-based cohorts to medico-administrative databases of the National health insurance (SNIIRAM database). In 2013, participants were asked in an ad-hoc questionnaire whether they wish to provide their social security number or not. Medical data regarding 50,240 participants of the NutriNet-Santé study were linked to the insurance database. Thus, for all these individuals, medical data were completed by the additional information from these databases, thereby limiting potential bias due to people with CVD who may not report their disease to the study investigators. Since June 2013, new participants included in the cohort are automatically

24

asked to provide their social security number in the baseline questionnaire or can provide it later on during follow-up, which increased acceptance rate of the procedure to about 60% (versus about 20% with pending authorization and 20% participants who declined the linkage). This system is very recent and corresponds to deep changing in the access to medico-administrative data in France for research purposes, to which our team contributed as a pioneer.

Importantly, we have verified that results remained similar when analyses included only cases and censored participants with linked SNIIRAM data (HRfor a 10-point increment in the proportion of ultra-processed

foods in the diet=1.13 (1.06-1.12) p=0.0003 for CVD risk).

Besides, although SNIIRAM databases are conceptually interesting, they allowed us to detect only 9 missing cardiovascular events, showing that our follow-up procedure in the NutriNet-Santé data is already performant. The SNIIRAM database theoretically covers all reimbursements of healthcare and hospitalizations by the social security. However, these data were not initially designed for research purposes and errors may be found in these databases: we have detected some cases reported in our questionnaires and validated on the basis of medical documents by our physicians, but which were not in the SNIIRAM databases (n = 23). The combination of these different collection methods for case ascertainment therefore maximizes case coverage, although exhaustiveness can never be guaranteed. Regarding deceased participants, medical information is collected using the SNIIRAM databases if a prior authorization has been given by the participant. Otherwise, the family and/or treating physician are contacted, and eventually, missing medical information is completed with an additional and exhaustive linkage to the French National cause-specific mortality registry (CépiDC, which includes both dates and causes of death, and is accessible for all French citizens, without specific authorization or identification number).”

Additional information has also been added to the manuscript regarding the following aspects upon Reviewer 5’s request:We have calculated incidence of CVD according to the number of completed dietary record. Subjects with ≤6 records had an incidence rate of 209 cases for 100,000 person-years (mean age 40.6y), compared to 344 for 100,000 person-years in subjects with >6 records (mean age 46.6 y); however, similar results were observed in both groups of participants: respectively HRfor a 10-

point increment in the proportion of ultra-processed foods in the diet= 1.13 (1.03 to 1.24), p=0.009, and HR = 1.11 (1.01 to 1.23), p=0.03. We have added these results section, page 9.A very low proportion of participants (1.7%) emigrated to other countries and were not covered by the SNIIRAM database. We have added this precision in the Methods page 6.

4. As mentioned in the text, the study population is a volunteer-based sample. What risk factors of CVDs were differed from the target population? And how did these factors influence the association and generalizability of the finding. Is there any difference in the age- and sex-standardized incidence of CVDs between the study population and the source population according to the SNIIRAM databases? More detailed discussion is needed. --> Indeed, participants included in NutriNet-Santé were volunteers who agreed to participate to long-term studies on lifestyle factors and health. By construction, recruited participants are

25

not representative of the general population of the country. This is the case for most large-scale population-based cohorts worldwide (including the highly prestigious ones such as EPIC, or the Harvard’s Nurses’ Health Study and Health Professional Follow Up Study). This is not specific to the NutriNet-Santé cohort. In such etiological studies representativeness is less crucial than diversity of exposure among studied populations (please see our response to Reviewer 2, page 7 of this document). However, we agree that this may introduce some bias in HR estimates that should be discussed. Compared to the general French population, participants to the NutriNet-Santé cohort were younger, more often women, with health-conscious behaviours and higher socio-professional and educational levels (2). There were less smokers in the NutriNet-Santé cohort (14), less individuals with overweight/obese (28.2% in men and 29.4% in women in NutriNet-Santé vs. 54% in men and 44% in women in the French population) (15), and less type 2 diabetic patients (baseline prevalence in the cohort = 1.6% versus 6% in the French population (16)). Participants of the NutriNet-Santé cohort also had healthier dietary intakes than the French population: they consumed more fruit and vegetables and fish, and less meat and added fats (14). This may have resulted in 1) a lower CVD incidence compared to national estimates (age and sex-standardized incidence rate per 100,000 persons per year: 495 cases in our cohort vs 500 in France (17), although these figures are not strictly comparable because unlike in our cohort, no national data is available for non-hospitalized TIA in France) and 2) an overall lower exposure to ultra-processed foods, with less contrast between extreme categories (no available national data in France). These points rather tended to underestimate the strength of the associations. However, the possibility that selection bias may have led to an overestimation of some associations cannot be totally excluded. We have added some discussion about this point in the limits section of the manuscript, page 13.

5. In cohort studies, some feedbacks or advices are needed for the participants at each data collection. The participants’ dietary behaviors may thus be changed due to the relevant feedbacks or advices. To clarify the influence of this issue, it is a need to conduct a sensitivity analysis by limited to those participants with the similar proportion (e.g., the Z-score of the changes in proportion ranged -1 to 1) of ultra-processed foods during the whole follow-up period. --> In this observational study, in order to avoid any modification of dietary behaviors, no individual data or advice is transmitted to the participants (only general information on scientific results from the study). Moreover, the ultra-processed food topic is very recent in France for the general public, thus dietary modification regarding this specific aspect is of low probability for the dietary data assessment period considered. However, in order to comply with Reviewer 5’s request, we have now performed a sensitivity analysis by focusing on the participants for whom the proportion of ultra-processed foods in the diet varied by less than │10%│ between the beginning and the end of their follow-up, which provided similar results (1029 CVD cases and 84203 non-cases, HRfor a 10-point increment in the proportion of ultra-processed foods in the

diet=1.09 (1.00 to 1.19), P=0.04). These points have been added in the Methods section page 8 and the Results section page 10.

Minor comments

26

6. Although the study size is large, the sample size is not large enough to obtain a precise association (particular for the individual types of CVDs) due to limited follow-up time. So, the strength of the large sample size is not valid in the discussion. --> We have removed this statement from the strengths in the discussion, page 12. 7. Appendix 7, the findings showed that the total association was mainly accounted for the subtypes of beverages, fats and sauces, meat, fish and egg, sugary products and salty snacks. In the section of title, discussion and conclusions, only “ultra-processed foods” were mentioned. It might be misunderstood to all ultra-processed foods. It is needed to discuss this issue the Discussion section. --> In this study, indeed, the strongest associations were observed for ultra-processed beverages, fats and sauces, meat, fish and eggs, sugary products and salty snacks. However, as explained above: “The effects of ultra-processed foods on human health may go through complex mechanisms involving synergic effects of many compounds and characteristics of ultra-processed foods. A chronic exposure to multiple factors, including cocktails of food additives (e.g. glutamate salts frequently used in sauces, artificial sweeteners frequently used in beverages, preservatives frequently used in ready-to-eat meals, etc…), neoformed compounds and contact materials may play a role in the studied association. These mechanisms can hardly be dispatched based on food groups as they should be considered globally. Creating an indicator for the proportion of ultra-processed foods in the diet allows distinguishing individuals with a high/low exposure to these cocktail interactions. The fact that the associations were stronger when considering the overall ultra-processed food proportion in the diet, rather than the associations in specific food groups argue in favor of these potential cocktail effects. A key perspective of the present study will be to go further in the understanding of underlying mechanisms and potentially deleterious factors. In particular, our research team is currently launching a large-scale epidemiological and experimental programme to investigate chronic exposure to food additives (single substances and multi-exposure cocktails) and its potential impacts on health (13).”We have added these precisions to the discussion pages 12-13.

8. Table 2, a crude or minimumly adjusted model is usually required. --> A crude model (Model 0) only adjusted for sex (+age as time scale) has been added to Table 2.

9. Page 9, line 15: A whole range or interquartile range of follow-up time might help the readers to understand the cohort. --> Interquartile range of follow-up time has been added, page 9.

10. Page 9, line 16: TIA accounted for 47.8% of the total CVD events in this study. Usually, TIA is the lightest CVD events, and there is no strong objective evidence for the TIA diagnosis. If TIA can be included as a CVD event, why not included angina? Was the proportion rate of CVD events in the study population comparable to those in the source population? More details are needed. --> TIA corresponds to a brief episode of neurological dysfunction that has the same underlying mechanism as ischemic stroke. TIA and stroke share common causes (disruption of cerebral

27

blood flow) and the same symptoms, even though TIA lasts for a shorter time and usually leaves no evidence of infarction on imaging. It might often lead to hospitalization, especially when recurrent. Thus, we have included this CVD in the main model, but we also tested analyses excluding TIA, which provided similar results (HRfor a 10-point increment in the proportion of ultra-processed foods in

the diet=1.12 (1.02 to 1.23), P=0.02, 754 cases and 104405 non cases) (page 9). Concerning angina: angina pectoris events were already included as ACS (ICD code I20) in this study. In contrast, stable angina events are soft CVD events occurring only during effort or intense physical activity. They usually do not require hospitalization, and the first occurrence of a stable angina might not always be an alarming signal for the patient, so it might be missed. Moreover, they might have other causes than coronary obstruction, like anemia, abnormal heart rhythms and heart failure. For these reasons, we did not include stable anginas in the main model. However, in order to comply with Reviewer’s 5 comment, we have now tested the analyses including stable angina events. Results remained unchanged ((HRfor a 10-point increment in the

proportion of ultra-processed foods in the diet =1.12 (1.06 to 1.19), P=0.0002, 1601 cases and 103120 non-cases). We have now added this analysis in the manuscript in the methods and results section on pages 8 and 9.French national data regarding CVD events include only hospitalizations by CVD, thus they do not take into account non-hospitalized TIAs for instance, making the comparison with our data not straightforward. Age and sex standardized incidence rate per 100000 persons per year were 495 CVD cases in our cohort vs 500 in France (17). This information has now been added to the manuscript on page 13. Additional Questions:Please enter your name: Yuming ChenJob Title: ProfessorInstitution: Sun Yat-sen UniversityReimbursement for attending a symposium?: NoA fee for speaking?: NoA fee for organising education?: NoFunds for research?: NoFunds for a member of staff?: NoFees for consulting?: NoHave you in the past five years been employed by an organisation that mayin any way gain or lose financially from the publication of this paper?: NoDo you hold any stocks or shares in an organisation that may in any waygain or lose financially from the publication of this paper?: NoIf you have any competing interests <A HREF='http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/declaration-competing-interests'target='_new'> (please see BMJ policy) </a>please declare them here: None.

Reviewer: 6

Recommendation:

28

Comments:The authors of the manuscript assessed the association between ultra-processed food and risk of cardiovascular diseases in a French cohort, and reported that high consumption of ultra-processed food was associated with a higher risk of all cardiovascular disease outcomes. The topic is current, and the manuscript is well-written overall, but some issues should be addressed. --> We sincerely thank Reviewer 6 for this positive comment and for her valuable insights. All issues pointed out by Reviewer 6 were addressed below.

Major comments:

1) One overarching concern about the study is that ultra-processed food is a very heterogeneous exposure. Shop bought whole grain bread for example, if I understood correctly, would be regarded as ultra-processed, although it seems unconvincing that it would be associated with disease risk in the same manner as some other ultra-processed foods such as sugar sweetened beverages or processed meat. Indeed, when the authors assessed the risk for individual foods (Appendix 7), there was no significant association for many foods. In light of this, the results should be interpreted with caution, and the resulting public health message is unclear. --> We fully agree with Reviewer 6 that ultra-processed foods represent a large spectrum of food products. However, in this study, some associations were observed for many ultra-processed food groups (beverages, fats and sauces, meat, fish and eggs, sugary products and salty snacks). Most importantly, as explained above: “The effects of ultra-processed foods on human health may go through complex mechanisms involving synergic effects of many compounds and characteristics of ultra-processed foods. A chronic exposure to multiple factors, including cocktails of food additives (e.g. glutamate salts frequently used in sauces, artificial sweeteners frequently used in beverages, preservatives frequently used in ready-to-eat meals, etc…), neoformed compounds and contact materials may play a role in the studied association. These mechanisms can hardly be dispatched based on food groups as they should be considered globally. Creating an indicator for the proportion of ultra-processed foods in the diet allows distinguishing individuals with a high/low exposure to these cocktail interactions. The fact that the associations were stronger when considering the overall ultra-processed food proportion in the diet, rather than the associations in specific food groups argue in favor of these potential cocktail effects. A key perspective of the present study will be to go further in the understanding of underlying mechanisms and potentially deleterious factors. In particular, our research team is currently launching a large-scale epidemiological and experimental programme to investigate chronic exposure to food additives (single substances and multi-exposure cocktails) and its potential impacts on health (13).” This information has been added pages 12-13.In the meanwhile, in terms of public health message, even if it remains uncertain to date which specific processes, compounds or ultra-processed food subtypes play a more important role, evidence is accumulating for an association between increased overall proportion of ultra-processed food in the diet and increased risks of several chronic diseases such as metabolic syndrome, dyslipidemia, hypertension, cancer, or gastrointestinal disorders (18–22). It is

29

therefore important to inform the consumers about these associations, and to recommend reducing the proportion of ultra-processed food in the diet and promoting the consumption of unprocessed/minimally processed foods instead, as it is already the case in official nutritional recommendations in France and Brazil, in the name of the precautionary principle (23,24). Industrials should also be encouraged to improve the quality of their products, notably by enhancing nutritional composition and reducing the use of unnecessary additives. We have added some precisions in the conclusion page 14.

2) Diet was assessed using 24 hour recalls collected within the first two years of follow-up, and the values were averaged. This could result in a situation where the exposure was assessed after the outcome, if the participant had an event within the first year for example. Excluding the first two years of follow-up would therefore seem insufficient to assess the likelihood of reverse causality, as a participant might change their diet close to the end of the initial two year period, which would influence their CVD risk in the few years after that. A longer exclusion (e.g. 5 years) would appear necessary, and in my view, it would be more appropriate to show the main results by excluding cases within the first two years, to avoid violating the temporality principle.--> In order to answer to Reviewer’s 6 question, we have strengthened sensitivity analyses to challenge our data regarding potential reverse causality. We previously showed that findings were similar when cases diagnosed during the first two years were excluded. This analysis is now presented in the Results section of the main manuscript page 10. We additionally tested the exclusion of cases during the first three, four, and five years of follow-up and results remained statistically significant. In order to keep maximal statistical power for our main analyses, we kept all cases in our main model, but a full paragraph has been added to the Results section page 10 which now reads: “The associations remained significant after the exclusion of CVD cases diagnosed during the first two years of follow-up: HR=1.14 (1.05 to 1.23), P=0.0008, 1087 cases and 103750 non cases (appendix 9), as well as during the first three (HR=1.44 (1.05 to 1.25), P=0.002, 879 cases and 103750 non cases), four (HR=1.44 (1.03 to 1.25), P=0.01, 663 cases and 103750 non cases) and five years (1.13 (1.00 to 1.28), P=0.04, 441 cases and 103750 non cases).”As usually done in nutritional epidemiology, the assumption is made that the measured exposure at baseline (especially since we averaged a two-year period of exposure) actually reflects more generally the usual eating habits of the individual during adulthood, including several years prior to his/her entry into the cohort, and several years after. Besides, in a prior cross-sectional study in the NutriNet-Santé cohort (25) we showed that participants with cardiovascular diseases had similar alcohol and tobacco consumption and did not have healthier dietary behaviours compared to healthy participants, which did not support a drastic change towards healthier dietary habits after a CVD event. Finally, as also shown in our response to Reviewer 5 point 5, results were stable when another approach for temporality of dietary data was considered, i.e. focusing on the participants for whom the proportion of ultra-processed foods in the diet varied by less than │10%│ between the beginning and the end of their follow-up (page 10).

30

3) The NutriNet-Sante cohort consists of relatively young participants, where CVD is perhaps less common compared to the general population? Absolute risk estimates in addition to the currently reported relative risks would give a better indication of the level of baseline risk as well as risk difference and population impact. --> Indeed, the NutriNet-Santé consists of younger participants than the French national population (2) with lower CVD incidence. As explained above in our response to Reviewer 5 point 4, although a straightforward comparison is not possible due to lack of French national data for non-hospitalized CVD, we have added age and sex-standardized CVD incidence rate per 100000 persons per year to the manuscript on page 13 : 495 cases in our cohort (253 without standardization) vs 500 in France (17). In order to comply with Reviewer’s 6 request, we have also added absolute age and sex-corrected risk estimates to the Results section page 9: “Absolute CVD risks were 253 for 100000 person years in the whole population: more specifically, age and sex-corrected absolute CVD risks were 242 for 100000 person years in the first quartile of the proportion of ultra-processed food intake in the diet, 254 in the second, 252 in the third and 277 in the fourth quartile.”

4) The authors reported that the proportion of unprocessed/minimally processed foods was inversely associated with CVD risk. Could the higher risk associated with ultra-processed food be explained by the simultaneous low consumption of unprocessed/minimally processed food instead? Did the authors test for a model adjusting for minimally processed food consumption?--> We thank Reviewer 6 for this insightful comment that allows us to precise certain points. Our main exposure variable is expressed as a percentage that quantifies the proportion of ultra-processed foods in the whole diet. By construction, people having an overall higher share of ultra-processed foods in their diets also had an overall lower proportion of non-ultra-processed foods. More specifically, the proportion of ultra-processed and of unprocessed/minimally processed food in the diet are almost collinear (Person correlation coefficient=-0.8). Thus, adjustment for non-ultra-processed food intake in the model is not methodological feasible since it would represent a strong over-adjustment. As a matter of fact, we showed in a secondary analysis page 10 that, consistently, a higher proportion of unprocessed/minimally foods was associated with lower risks of CVD.Thus part of the association between ultra-processed food intake and CVD risk probably goes indeed through the simultaneous (by construction) lower consumption of non-ultra-processed foods. However, this does not explain the whole association since ultra-processed food may also exert more direct effects though complex mechanisms notably involving synergic effects of food additives cocktails, neoformed compounds and contact materials. This was supported by our results showing that several ultra-processed food groups were associated with increased CVD risk while the non-ultra-processed form of these food groups were not, as explained in our response to Reviewer 5 point 1 (now added to the manuscript page 9 and Appendix 8-b).The consequence in terms of public health messages is that, in light of these results and of mounting evidence from the literature regarding associations between the proportion of ultra-processed foods in the diet and several chronic disease risks, recommendations should be to limit the proportion of ultra-processed foods in the diet and promote the consumption of unprocessed/minimally processed foods instead.

31

This has been clarified in the manuscript: What this study adds box (page 2), abstract (page 3), and discussion (page 11).

Minor comments:Abstract5) The average number of repeated 24 hour dietary record could be mentioned in the settings and participants section.--> The average number of repeated 24 hour dietary records has now been added to the abstract.

6) Both in the abstract, main results and tables, mean follow-up and person-years for each outcome should be mentioned. --> As requested, mean follow-up and person-years have been added in the abstract, main results, and tables, throughout the manuscript.

Methods7) Relevant ICD codes could be added for each outcome. --> ICD codes for each outcome have been added in the Methods section pages 6 and 7.

Additional Questions:Please enter your name: Tammy TongJob Title: Nutritional EpidemiologistInstitution: University of OxfordReimbursement for attending a symposium?: NoA fee for speaking?: NoA fee for organising education?: NoFunds for research?: NoFunds for a member of staff?: NoFees for consulting?: NoHave you in the past five years been employed by an organisation that mayin any way gain or lose financially from the publication of this paper?: NoDo you hold any stocks or shares in an organisation that may in any waygain or lose financially from the publication of this paper?: NoIf you have any competing interests <A HREF='http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/declaration-competing-interests'target='_new'> (please see BMJ policy) </a>please declare them here:

**Information for submitting a revision**

Deadline: Your revised manuscript should be returned within one month.

32

How to submit your revised article: Log into http://mc.manuscriptcentral.com/bmj and enter your Author Center, where you will find your manuscript title listed under "Manuscripts with Decisions." Under "Actions," click on "Create a Revision." Your manuscript number has been appended to denote a revision.

You will be unable to make your revisions on the originally submitted version of the manuscript. Instead, revise your manuscript using a word processing program and save it on your computer. Once the revised manuscript is prepared, you can upload it and submit it through your Author Center. When submitting your revised manuscript, you will be able to respond to the comments made by the reviewer(s) and Committee in the space provided. You can use this space to document any changes you make to the original manuscript and to explain your responses. In order to expedite the processing of the revised manuscript, please be as specific as possible in your response to the reviewer(s). As well as submitting your revised manuscript, we also require a copy of the manuscript with changes highlighted. Please upload this as a supplemental file with file designation ‘Revised Manuscript Marked copy’. Your original files are available to you when you upload your revised manuscript. Please delete any redundant files before completing the submission.

When you revise and return your manuscript, please take note of all the following points about revising your article. Even if an item, such as a competing interests statement, was present and correct in the original draft of your paper, please check that it has not slipped out during revision. Please include these items in the revised manuscript to comply with BMJ style (see: http://www.bmj.com/about-bmj/resources-authors/article-submission/article-requirements andhttp://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists).

Items to include with your revision (see http://www.bmj.com/about-bmj/resources-authors/article-types/research):

1. What this paper adds/what is already known box (as described at http://resources.bmj.com/bmj/authors/types-of-article/research)

2. Name of the ethics committee or IRB, ID# of the approval, and a statement that participants gave informed consent before taking part. If ethics committee approval was not required, please state so clearly and explain the reasons why (see http://resources.bmj.com/bmj/authors/editorial-policies/guidelines.)

3. Patient confidentiality forms when appropriate (see http://resources.bmj.com/bmj/authors/editorial-policies/copy_of_patient-confidentiality).

4. Competing interests statement (see http://resources.bmj.com/bmj/authors/editorial-policies/competing-interests)

33

5. Contributorship statement+ guarantor (see http://resources.bmj.com/bmj/authors/article-submission/authorship-contributorship)

6. Transparency statement: (see http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/transparency-policy)

7. Copyright statement/licence for publication (see http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/copyright-open-access-and-permission-reuse)

8. Data sharing statement (see http://www.bmj.com/about-bmj/resources-authors/article-types/research)

9. Funding statement and statement of the independence of researchers from funders (see http://resources.bmj.com/bmj/authors/article-submission/article-requirements).

10. Patient involvement statement (see http://www.bmj.com/about-bmj/resources-authors/article-types/research).

11. Please ensure the paper complies with The BMJ’s style, as detailed below:

a. Title: this should include the study design eg "systematic review and meta-analysis.”b. Abstract: Please include a structured abstract with key summary statistics, as explained below (also see http://resources.bmj.com/bmj/authors/types-of-article/research). For every clinical trial - and for any other registered study- the last line of the abstract must list the study registration number and the name of the register.c. Introduction: This should cover no more than three paragraphs, focusing on the research question and your reasons for asking it now.d. Methods: For an intervention study the manuscript should include enough information about the intervention(s) and comparator(s) (even if this was usual care) for reviewers and readers to understand fully what happened in the study. To enable readers to replicate your work or implement the interventions in their own practice please also provide (uploaded as one or more supplemental files, including video and audio files where appropriate) any relevant detailed descriptions and materials. Alternatively, please provide in the manuscript urls to openly accessible websites where these materials can be found.

e. Results: Please report statistical aspects of the study in line with the Statistical Analyses and Methods in the Published Literature (SAMPL) guidelines http://www.equator-network.org/reporting-guidelines/sampl/. Please include in the results section of your structured abstract (and, of course, in the article's results section) the following terms, as appropriate:

34

i. For a clinical trial: Absolute event rates among experimental and control groups; RRR (relative risk reduction); NNT or NNH (number needed to treat or harm) and its 95% confidence interval (or, if the trial is of a public health intervention, number helped per 1000 or 100,000.)ii. For a cohort study: Absolute event rates over time (eg 10 years) among exposed and non-exposed groups; RRR (relative risk reduction.)iii. For a case control study:OR (odds ratio) for strength of association between exposure and outcome.iv. For a study of a diagnostic test: Sensitivity and specificity; PPV and NPV (positive and negative predictive values.)v. For a systematic review and/or meta-analysis: Point estimates and confidence intervals for the main results; one or more references for the statistical package(s) used to analyse the data, eg RevMan for a systematic review. There is no need to provide a formal reference for a very widely used package that will be very familiar to general readers eg STATA, but please say in the text which version you used. For articles that include explicit statements of the quality of evidence and strength of recommendations, we prefer reporting using the GRADE system.f. Discussion: To minimise the risk of careful explanation giving way to polemic, please write the discussion section of your paper in a structured way. Please follow this structure: i) statement of principal findings of the study; ii) strengths and weaknesses of the study; iii) strengths and weaknesses in relation to other studies, discussing important differences in results; iv) what your study adds (whenever possible please discuss your study in the light of relevant systematic reviews and meta-analyses); v) meaning of the study, including possible explanations and implications for clinicians and policymakers and other researchers; vi) how your study could promote better decisions; vi) unanswered questions and future researchg. Footnotes and statements

Online and print publication: All original research in The BMJ is published with open access. Our open access policy is detailed here: http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/copyright-open-access-and-permission-reuse. The full text online version of your article, if accepted after revision, will be the indexed citable version (full details are at http://resources.bmj.com/bmj/about-bmj/the-bmjs-publishing-model).END--> This editorial checklist has been carefully followed and the manuscript has been edited accordingly. We thank you again for your interest in our work.

35

--> References

1. Desquilbet L, Mariotti F. Dose-response analyses using restricted cubic spline functions in public health research. Stat Med. 2010 Apr 30;29(9):1037–57.

2. Andreeva VA, Salanave B, Castetbon K, Deschamps V, Vernay M, Kesse-Guyot E, et al. Comparison of the sociodemographic characteristics of the large NutriNet-Sante e-cohort with French Census data: the issue of volunteer bias revisited. JEpidemiolCommunity Health. 2015 Sep;69:893–8.

3. Black AE. Critical evaluation of energy intake using the Goldberg cut-off for energy intake:basal metabolic rate. A practical guide to its calculation, use and limitations. IntJObesRelat Metab Disord. 2000 Sep;24:1119–30.

4. Black AE. The sensitivity and specificity of the Goldberg cut-off for EI:BMR for identifying diet reports of poor validity. Eur J Clin Nutr. 2000 May;54(5):395–404.

5. Goldberg GR, Black AE, Jebb SA, Cole TJ, Murgatroyd PR, Coward WA, et al. Critical evaluation of energy intake data using fundamental principles of energy physiology: 1. Derivation of cut-off limits to identify under-recording. Eur J Clin Nutr. 1991 Dec;45(12):569–81.

6. Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr. 1985;39 Suppl 1:5–41.

7. Lassale C, Castetbon K, Laporte F, Camilleri GM, Deschamps V, Vernay M, et al. Validation of a Web-based, self-administered, non-consecutive-day dietary record tool against urinary biomarkers. BrJNutr. 2015 Mar 28;113(1475-2662 (Electronic)):953–62.

8. DiMeglio DP, Mattes RD. Liquid versus solid carbohydrate: effects on food intake and body weight. Int J Obes. 2000 Jun;24(6):794–800.

9. Takeuchi M, Takino J, Furuno S, Shirai H, Kawakami M, Muramatsu M, et al. Assessment of the Concentrations of Various Advanced Glycation End-Products in Beverages and Foods That Are Commonly Consumed in Japan. PLOS ONE. 2015 Mar 2;10(3):e0118652.

10. Uribarri J, Stirban A, Sander D, Cai W, Negrean M, Buenting CE, et al. Single Oral Challenge by Advanced Glycation End Products Acutely Impairs Endothelial Function in Diabetic and Nondiabetic Subjects. Diabetes Care. 2007 Oct 1;30(10):2579–82.

11. Bingham SA, Gill C, Welch A, Day K, Cassidy A, Khaw KT, et al. Comparison of dietary assessment methods in nutritional epidemiology: weighed records v. 24 h recalls, food-frequency questionnaires and estimated-diet records. Br J Nutr. 1994 Oct;72(4):619–43.

12. Prentice RL, Mossavar-Rahmani Y, Huang Y, Van Horn L, Beresford SAA, Caan B, et al. Evaluation and Comparison of Food Records, Recalls, and Frequencies for Energy and

36

Protein Assessment by Using Recovery Biomarkers. Am J Epidemiol. 2011 Sep 1;174(5):591–603.

13. Gourd E. Ultra-processed foods might increase cancer risk. Lancet Oncol. 2018 Apr 1;19(4):e186.

14. Andreeva VA, Deschamps V, Salanave B, Castetbon K, Verdot C, Kesse-Guyot E, et al. Comparison of Dietary Intakes Between a Large Online Cohort Study (Etude NutriNet-Santé) and a Nationally Representative Cross-Sectional Study (Etude Nationale Nutrition Santé) in France: Addressing the Issue of Generalizability in E-Epidemiology. Am J Epidemiol. 2016 Nov 1;184(9):660–9.

15. Santé Publique France. ESTEBAN - A health study on environment, biosurveillance, physical activity and nutrition 2014-2016 (Original title in French: Etude de santé sur l’environnement, la biosurveillance, l’activité physique et la nutrition (Esteban) [Internet]. 2017 [cited 2018 Oct 17]. Available from: http://invs.santepubliquefrance.fr/Publications-et-outils/Rapports-et-syntheses/Environnement-et-sante/2017/Etude-de-sante-sur-l-environnement-la-biosurveillance-l-activite-physique-et-la-nutrition-Esteban-2014-2016

16. Santé Publique France. Prevalence and incidence of diabetes (in French) / Epidemiological data [Internet]. [cited 2018 Oct 17]. Available from: http://invs.santepubliquefrance.fr/Dossiers-thematiques/Maladies-chroniques-et-traumatismes/Diabete/Donnees-epidemiologiques/Prevalence-et-incidence-du-diabete

17. Santé Publique France, DREES. Health situation of the French Population in 2017 (Original title in French: L’état de santé de la population en France, Rapport 2017) [Internet]. [cited 2018 Oct 18]. Available from: http://invs.santepubliquefrance.fr/publications/etat_sante_2017/ESP2017_Ouvrage_complet_vdef.pdf

18. Fiolet T, Srour B, Sellem L, Kesse-Guyot E, Allès B, Méjean C, et al. Consumption of ultra-processed foods and cancer risk: results from NutriNet-Santé prospective cohort. BMJ. 2018 14;360:k322.

19. Rauber F, Campagnolo PD, Hoffman DJ, Vitolo MR. Consumption of ultra-processed food products and its effects on children’s lipid profiles: a longitudinal study. NutrMetab Cardiovasc. 2015 Jan;25(1590-3729 (Electronic)):116–22.

20. Mendonca RD, Pimenta AM, Gea A, Fuente-Arrillaga C, Martinez-Gonzalez MA, Lopes AC, et al. Ultraprocessed food consumption and risk of overweight and obesity: the University of Navarra Follow-Up (SUN) cohort study. AmJClinNutr. 2016 Nov;104(1938-3207 (Electronic)):1433–40.

21. Mendonca RD, Lopes AC, Pimenta AM, Gea A, Martinez-Gonzalez MA, Bes-Rastrollo M. Ultra-Processed Food Consumption and the Incidence of Hypertension in a Mediterranean

37

Cohort: The Seguimiento Universidad de Navarra Project. AmJHypertens. 2017 Apr 1;30(1941-7225 (Electronic)):358–66.

22. Schnabel L, Buscail C, Sabate J-M, Bouchoucha M, Kesse-Guyot E, Allès B, et al. Association Between Ultra-Processed Food Consumption and Functional Gastrointestinal Disorders: Results From the French NutriNet-Santé Cohort. Am J Gastroenterol. 2018 Jun 15;

23. Ministry of Health of Brazil. Dietary Guidelines for the Brazilian population [Internet]. 2014. Available from: http://189.28.128.100/dab/docs/portaldab/publicacoes/guia_alimentar_populacao_ingles.pdf

24. Haut Conseil de la santé publique. Avis relatif à la révision des repères alimentaires pour les adultes du futur Programme national nutrition santé 2017-2021. 2017 p.

25. Adriouch S, Lelong H, Kesse-Guyot E, Baudry J, Lampuré A, Galan P, et al. Compliance with Nutritional and Lifestyle Recommendations in 13,000 Patients with a Cardiometabolic Disease from the Nutrinet-Santé Study. Nutrients. 2017 May 26;9(6).