NutriNet-Brasil, a web-based prospective study on dietary patterns and risk of chronic diseases: cohort profile

Abstract

Purpose

Non-communicable diseases (NCDs), such as diabetes, cardiovascular diseases and cancer, are major global public health concerns. Diet quality—particularly the consumption of ultra-processed foods—has been associated with increased risk of NCDs. Traditional cohort studies are often expensive and logistically complex. The NutriNet-Brasil cohort leverages a web-based approach, offering a cost-effective and practical solution for comprehensive data collection and long-term follow-up.

Participants

Recruitments began in January 2020 through mass media, social media campaigns and collaborations with health organisations. Eligible participants are adults (aged ≥18 years) living in Brazil with internet access. Participants complete self-administered online questionnaires covering dietary intake, health status and other health determinants. Dietary assessment is based on the Nova classification system, which categorises foods by their level of processing.

Findings to date

Over 88 000 participants have completed the initial questionnaire. The cohort is predominantly women (79.9%) and highly educated (67.9% had completed higher education). The web-based design enabled the development and application of innovative dietary assessment tools, including the Nova24h and the Nova24hScreener, specifically designed to evaluate food processing levels. These tools have shown good performance in capturing dietary patterns and are central to the cohort’s aim. The online platform facilitates efficient recruitment, data collection and participant retention.

Future plans

NutriNet-Brasil is pioneering the development of web-based cohort methodologies and instruments tailored to food processing research. Future work includes leveraging collaborations with national and international research centres to conduct multidisciplinary analyses and inform public health policies.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This is the first nationwide cohort in Brazil developed entirely through an online platform, enabling rapid, cost-effective data collection without the need for face-to-face interviews.

• The prospective design and large sample size enhance the study’s ability to investigate temporal associations between dietary patterns and health outcomes.

• Although the sample is not representative of the overall adult population in Brazil, selection bias is unlikely to materially affect the validity of associations between dietary patterns and health outcomes.

• Innovative self-reported tools, including the Nova24h and the Nova24hScreener, were developed specifically to assess food consumption according to food processing.

• Despite reliance on self-reported data for diet, lifestyle and health conditions, previous validation studies have shown good agreement between web-based and objectively measured data.

Introduction

Non-communicable diseases (NCDs), such as diabetes, cardiovascular diseases and cancer, are major global public health concerns and represent challenges for health policies. In 2019, NCDs accounted for 1.62 billion age-standardised disability-adjusted life years, increasing from 43.2% in 1990 to 63.8% in 2019.¹ This rise is attributed to increasing exposure to metabolic risk factors, such as high body mass index (BMI), elevated fasting plasma glucose and high systolic blood pressure, alongside behavioural factors, including unhealthy diets, smoking and physical inactivity.² While the pivotal role of diet quality in preventing NCDs is well established,³ studies have shifted from a reductionist focus on individual foods or nutrients to recognising the importance of dietary patterns.⁴,⁶ Additionally, accumulating evidence suggests that the global shift from traditional, local diets to ultra-processed dietary patterns has substantially contributed to poor diet quality worldwide.⁷

Recent systematic reviews and meta-analyses from cohort studies, conducted mostly in Europe and the USA, have explored this paradigm shift under the food processing framework, consistently identifying associations between greater ultra-processed food (UPF) consumption and increased risk of NCDs.⁸,¹¹ However, some notable gaps persist. First, findings from these studies may not be generalisable to populations with distinct cultural dietary patterns. Second, traditional dietary assessment methods were not specifically designed to evaluate the degree of industrial food processing, highlighting the need for specific tools to capture this dimension accurately. Lastly, there is a scarcity of studies exploring different dietary patterns, such as plant-based versus animal-based diets, within the food processing framework, which could offer a more comprehensive understanding of the diet-NCD relationship.

To address these gaps, Brazilian researchers launched the NutriNet-Brasil Cohort Study, a large-scale, web-based study aiming to investigate the association between dietary patterns and health outcomes in Brazil, with particular focus on food processing. The project is coordinated by the Centre for Epidemiological Studies in Health and Nutrition at the University of São Paulo (Nupens/USP) in collaboration with national and international institutions. Recruitments began in January 2020 and currently enrol over 88 000 participants who completed the first questionnaire and will be followed up for at least 10 years (https://nutrinetbrasil.fsp.usp.br/).

The NutriNet-Brasil draws inspiration from the NutriNet-Santé Study initiated in France in 2009—the first national web-based cohort study specifically designed to study the association between diet and health.¹² This study has provided evidence on the association of various dietary exposures with NCDs, including specific food components and dietary patterns based on food processing.¹³,²⁰

Both cohorts capitalise on the global expansion of internet access, which has transformed the landscape of epidemiological research. In France, about 65% of the French population regularly used the internet in 2009. In Brazil, 92.5% reported having internet access in 2023, primarily through mobile devices—an increase of 1.0 percentage point from the previous year.²¹ These advances have facilitated the recruitment and follow-up of a large number of participants through online platforms, offering simpler logistics, faster data collection and lower operational costs.^{22 23}

This paper outlines the methodologies of the NutriNet-Brasil Cohort Study, providing a unique opportunity to examine the association between diet and health within the evolving context of globalised food systems and changing dietary cultures.

Cohort description

Objective and study design

The NutriNet-Brasil cohort aims to prospectively investigate the effect of dietary patterns and food processing on NCDs, including obesity, diabetes, hypertension, dyslipidaemia, cardiovascular diseases (eg, myocardial infarction and stroke), various types of cancer (eg, breast, lung and colorectal cancer), gastrointestinal disorders (eg, ulcerative colitis and Crohn’s disease), respiratory conditions (eg, emphysema and asthma), migraine, mental health disorders (eg, anxiety and depression), chronic kidney disease, liver diseases (including hepatic steatosis and cirrhosis), lupus, osteoporosis and thyroid disorders. This ongoing cohort relies on voluntary participation and is conducted entirely online, with data collected through self-administered digital questionnaires.

Eligibility and recruitment strategies

Eligible participants are adults (aged ≥18 years) living in Brazil with internet access. Recruitment began in January 2020 through mass media (especially television), endorsement from health and food organisations and digital dissemination strategies during the COVID-19 pandemic.

In December 2020, a large-scale social media campaign was launched, involving over 180 digital influencers from diverse regions and professional backgrounds (eg, food and cooking, health and science), which substantially expanded the study’s reach. Since December 2021, recruitment has been further supported through posters displayed in primary healthcare units across Brazil, in partnership with the Ministry of Health and municipal health departments.

Continuous recruitment is sustained through social media activities, leveraging peaks of enrolment following major media exposure (eg, television broadcasts and popular podcasts). Snowball sampling is also encouraged: participants are invited to recommend the study to friends and family on completing new questionnaires.

Data collection

Registration and consent form submission occur on the study’s digital platform, followed by an email sent for validation. At this point, participants provide their email, phone number, sex, state, city and date of birth. Once registration is validated, participants can access the survey page via login (email) and a personal password to complete the first questionnaire. New questionnaires are launched approximately every 3 months, designed to be brief to minimise participant burden. Notifications direct participants to the study platform, and questionnaires (except the first) must be completed within 60 days. Participants continue to be invited to future follow-ups regardless of missed responses. Figure 1 displays the timeline of the cohort.

Figure 1. Data collection timeline for the first 7 years after participants entered the NutriNet-Brasil cohort. Notes: participants are asked to self-report their weight in the 1st, 3rd and 10th months, as well as in the health status questionnaire (green boxes). Green and yellow boxes are always accompanied by the Nova24hScreener (short version). Red boxes are related to the Nova24h questionnaire (detailed version).

Participants respond to three types of questionnaires:

Dietary patterns assessment

Dietary patterns are assessed using two complementary tools that capture food consumption in the previous day, referred to in figure 1 as ‘short version’ and ‘detailed version.’ Both were validated in subsamples of the cohort. The detailed version is an open-ended, self-completed 24-hour food recall developed specifically for this cohort.²⁴ The Nova24h tool captures all the foods consumed in the previous day using probing questions to accurately differentiate the items based on the degree of industrial food processing. The Nova24h takes about 15 min to be completed, and it is first administered at the 7th-month questionnaire and once every 6 months thereafter (ie, the 11th-month and 17th-month questionnaires, and so on) (figure 1). The performance of Nova24h was assessed in a subsample of the cohort against an interviewer-led 24-hour recall in estimating the dietary contribution of the four Nova classification food groups. A total of 186 participants of the NutriNet-Brasil first completed the Nova24h and then received a telephone call from a well-trained dietitian for the application of a traditional 24-hour food recall, using the multiple-pass method for data collection.^{24 25} Intraclass correlation coefficients showed moderate to good agreement (0.54–0.78) between Nova24h and interviewer-led 24-hour food recall. Regarding the ability to rank participants according to quintiles of contribution of each Nova group, the study found substantial or almost perfect agreement between the tools (prevalence-adjusted and bias-adjusted kappa (PABAK) ranging from 0.69 to 0.81).²⁴

The Nova24hScreener is a 3 min self-administered food recall screener assessing the consumption (yes or no) of food items or food groups that are sentinel of unprocessed or minimally processed whole plant foods (WPFs) (33 items) and UPFs (23 items).²⁶ This screener is administered three times within the first 15 days after registration as part of the first questionnaire and subsequently along with each health status questionnaire (ie, the 14th-month, 20th-month and 26th-month questionnaires, and so on) (figure 1). The performance of two Nova diet quality scores derived from the Nova24hScreener in reflecting the consumption of unprocessed or minimally processed WPFs and UPFs (plant-based and animal-based) was assessed in a subsample of the cohort. Both scores (the Nova score of WPFs (Nova-WPFs) and the Nova score of UPFs (Nova-UPFs)) were compared with the dietary intake (percentage of total energy) of unprocessed or minimally processed WPFs and UPFs, respectively, obtained through the Nova24h, administered on the same day. There was a direct and linear relationship between the intervals of each score and the corresponding percentage of energy intake. Substantial agreement was found between the intervals of each score and the corresponding percentage of energy intake (PABAK 0.72, 95% CI 0.64 to 0.81 for the Nova-WPFs and PABAK 0.79, 95% CI 0.69 to 0.88 for the Nova-UPFs), suggesting that the two scores presented good performance in reflecting the intake of those food groups when obtained using a more complex and time-consuming tool.²⁶

Health and nutritional status assessment

Information on health status is collected based on medical diagnoses. Participants are first asked whether they had at least one medical consultation during the period between follow-ups (eg, ‘Did you visit a doctor since you completed your last health questionnaire?’). If they answer ‘yes,’ additional questions are presented, asking whether they received any specific diagnosis (eg, ‘Did a doctor diagnose you with any type of cancer?’). Each time a diagnosis is reported, that outcome will not be asked again in subsequent questionnaires, except for diagnoses that may occur more than once. The first questionnaire collects information on medical diagnoses and medication use related to hypertension, diabetes and hypercholesterolaemia. A comprehensive health status questionnaire is administered at the 14-month follow-up, including follow-up on these three conditions and baseline data on additional diseases, such as chronic respiratory diseases, cancers, cardiovascular diseases, stroke, chronic kidney disease, depression, migraine, irritable bowel syndrome, Crohn’s disease, ulcerative colitis and coeliac disease. This questionnaire also incorporates validated screening instruments for migraine,²⁷ depression,²⁸ irritable bowel syndrome symptoms,²⁹ anxiety³⁰ and cognitive decline³¹ (screeners are administered regardless of a doctor visit). During the COVID-19 pandemic, data on SARS-CoV-2 infection, symptom severity and hospitalisations were also collected. Further conditions were added at later follow-up points: lupus, osteoporosis and hepatic diseases at month 20 and thyroid disorders at month 38. The status of all health conditions is updated every six months following the initial assessment (ie, at months 20, 26, 32, 38, and so on), through questions about the incidence of new diseases since the last follow-up. Regarding nutritional status, height and weight are self-reported by participants in the first questionnaire. Weight is subsequently reassessed in all follow-up questionnaires, except those focused exclusively on dietary intake, allowing for updates at a minimum interval of 6 months.

Other determinants of health

Education level is assessed during the first questionnaire while other health determinants not related to food consumption are collected 3 months after registration. This information is updated at the 22nd month and subsequently every 2 years. These questionnaires address physical activity, sedentary behaviour, tobacco smoking, alcohol consumption and sleep habits, as well as electronic cigarette use. Information on reproductive health (women only) and detailed socioeconomic and demographic information (marital status, race/skin colour, occupation, assets, zip code of residence and Individual Taxpayer Registration number—in Portuguese, Cadastro de Pessoa Física (CPF)) is obtained only at the 10th questionnaire to avoid overwhelming participants in earlier questionnaires, which could lead to skipped responses or premature dropout.

The questions included in the NutriNet-Brasil questionnaires were mostly derived from previously validated instruments, such as the Global Physical Activity Questionnaire for assessing physical activity.³² For domains lacking validated instruments, questions from national health surveys, such as the Brazilian National Health Survey (Pesquisa Nacional de Saúde), and from the Longitudinal Study of Adult Health (Estudo Longitudinal da Saúde do Adulto), were adapted for self-administered online application.

Details about questionnaires and variables are presented in table 1.

Table 1. Questionnaires applied in the NutriNet-Brasil study and potential exposures or outcomes.

Categories	Questionnaires	Potential variables	Time points when questionnaire is applied
Dietary patterns	Food consumption on the previous day (short and detailed version)	Number of food groups consumed on the previous day; percentage of energy from food groups and macronutrients and micronutrients intake	Short version: first questionnaire, and subsequently along with each health status questionnaire. Detailed version: 7th-month questionnaire and once every 6 months thereafter.
Health and nutritional status	Health status	Weight/BMI changes. New cases of obesity, diabetes, hypertension, dyslipidaemia, cardiovascular diseases (such as myocardial infarction and stroke), various types of cancer (eg, breast, lung and colorectal cancer), gastrointestinal disorders (such as ulcerative colitis and Crohn’s disease), respiratory conditions (including emphysema and asthma), migraine, mental health disorders (such as anxiety and depression), chronic kidney disease, liver diseases (including hepatic steatosis and cirrhosis), lupus, osteoporosis and thyroid disorders.	Height and weight are self-reported by participants in the first questionnaire, and weight is subsequently reassessed in all follow-up questionnaires, except those focused exclusively on dietary intake. The first questionnaire collects information on medical diagnoses and medication use related to hypertension, diabetes and hypercholesterolaemia. A comprehensive health status questionnaire is administered at the 14th month. Lupus, osteoporosis and hepatic diseases at month 20 and thyroid disorders at month 38. The status of all reported conditions is updated every 6 months following their initial assessment.
Other life conditions	Life conditions and non-food consumption determinants of health status	Sociodemographic information (sex, age and residence location) and schooling level. Physical activity, sedentary behaviour, tobacco smoking, alcohol consumption, sleep habits and cognitive abilities. Reproductive health (women), marital status, race/skin colour, occupation, assets, zip code of residence and Individual Taxpayer Registration number.	Education level is assessed during the first questionnaire, while other life conditions are asked in a questionnaire 3 months after registration. This information is updated at the 22nd month and subsequently every 2 years. Information on reproductive health and detailed socioeconomic and demographic characteristics is obtained once at the 10th questionnaire.

BMI, Body Mass Index.

Linkage with nationwide official morbidity and mortality data

The full name of the participant’s mother is also requested in the Life Conditions Questionnaire (10th month). This information, along with the participant’s CPF and the date and locality of birth, will be used for future linkage with nationwide morbidity and mortality databases.³³ Sensitive participant information—specifically email, phone number, date of birth, CPF, mother’s name and ZIP code—is stored separately and is not available in the NutriNet-Brasil response databases. In these databases, participants are identified solely by an ID generated by the study platform. Hospital admissions, cancer cases and mortality data from official systems will also be linked to the cohort database. These linkages will be conducted by the Ministry of Health with the assistance of national research centres, such as the National Cancer Institute (INCA). The State Foundation for Data Analysis (SEADE) will provide support for the linkage with mortality data in the State of São Paulo.

Data quality control and diagnoses confirmation

The NutriNet-Brasil study questionnaires were designed to ensure data quality. For example, filters are employed to identify participants who did not complete a given question or questionnaire (eg, baseline of diseases, weight, height and sociodemographic characteristics) and to repeat the same enquiries in the next follow-up questionnaire only for those with missing values. Additionally, when participants are asked to enter their anthropometric data, only values within a predefined range are allowed (ie, for weight, the lower and upper limits are 25 and 300 kg, and for height, 1.00 and 2.40 m). When values outside these ranges are inserted or in case of missing values, anthropometric data are asked once again in the next follow-up questionnaire. For information on habitual sleep (average number of hours slept over the last 30 nights), there are also filters in the questionnaire that do not accept values below 3 hours or above 20 hours.

Participants self-reporting the diagnosis of a new condition from the 20th month onwards are invited to send any document that confirms the new diagnosis (ie, test result, doctor or hospital report or photo of the drug prescription or box prescribed to treat the disease). Diagnosis confirmation is sought for the following conditions: prediabetes, diabetes, hypercholesterolaemia, cancer, stroke, heart disease, gastrointestinal disease, respiratory disease, liver disease, lupus, osteoporosis, chronic renal failure and thyroid disease. Participants without these conditions are also invited to submit any test results from the past 6 months related to prediabetes, diabetes and hypercholesterolaemia, starting from the 26th-month questionnaire.

Information retrieved from the submitted documents is compiled into an ancillary database and used to confirm or refute the reported diseases. The NutriNet-Brasil study has a medical team to aid in interpreting documents and confirming diagnoses. This team contacts participants by telephone (with their permission) if needed. Confirmed diagnoses will be primarily used in sensitivity analyses.

Whenever an issue is identified in the questionnaires, they are promptly altered, prioritising greater clarity of the questions, minimising the burden on participants and enhancing the quality of the collected data.

Ethical aspects and data security

The NutriNet-Brasil study has obtained approval from the research ethics committee of the Faculty of Public Health at the USP (process no. 88455417.8.0000.5421) and has had its Data Management Plan sanctioned by the São Paulo Research Foundation (FAPESP) as part of funding 2022/05636–0. Monitoring of the study’s digital platform is carried out by an information technology team ensuring stringent security measures. Participant data is securely stored in a cloud infrastructure, safeguarding security and privacy concerns. Furthermore, the study’s platform adheres to the provisions of the Brazilian General Data Protection Law (LGPD), guaranteeing comprehensive privacy protection for all data housed within the study’s database, including the personal information of participants.

Researchers, research assistants and undergraduate or graduate students involved in data analysis have access only to datasets that have been stripped of sensitive participant information. Each study using NutriNet-Brasil data is registered through a project management system, and all individuals handling the data are required to sign a confidentiality agreement.

Findings to date

This paper describes the methods applied in the NutriNet-Brasil cohort, a web-based study that generates a unique data source to explore the effect of dietary patterns on NCDs in the Brazilian adult population, particularly in the context of evolving food systems and food processing trends. It presents comprehensive information on recruitment strategies, questionnaire administration and data quality control, serving as a valuable resource for researchers conducting similar Internet-based investigations.

The baseline period for analyses using NutriNet-Brasil data may include responses to multiple questionnaires, from registration to the point where exposure and outcome variables are established. Thus, the baseline sample composition varies according to the exposure and outcome assessed in each specific analysis. For instance, using the percentage of caloric contribution of UPFs obtained from at least two Nova24h recalls (needed to obtain usual intakes) must define the baseline at the 11th month (capturing 7th-month and 11th-month recalls).³⁴ In contrast, studies using the Nova24hScreener as an exposure can set the baseline at the third month.³⁵ For outcomes like diabetes, hypertension and hypercholesterolaemia—assessed early—the baseline can be defined close to enrolment. For diseases such as cancer, stroke and gastrointestinal disorders—first assessed at the 14th month—baseline must be established accordingly, with incident cases identified from the 20th month onwards.

Currently, the cohort includes approximately 88 000 participants who completed the first questionnaire. Longitudinal analyses require participants to have completed at least two dietary recalls (short or detailed) and the 14th-month health questionnaire. Table 2 describes this analytical sample (n=54 057) in terms of sociodemographic characteristics (obtained either at registration in the study or at the first questionnaire) and compares its sociodemographic characteristics with those of the general Brazilian adult population. Participants are predominantly female (79.9% vs 51% nationally), university-educated (67.9% vs 10.6%) and aged between 30 and 60 years. Regional representation is comparable to the national distribution in the Midwest (7.8% vs 7.3%) and South (15.9% vs 14.3%), but participants are over-represented from the Southeast (62.2% vs 42.1%) and under-represented from the Northeast (11.2% vs 17.3%) and North (2.9% vs 8.3%).³⁶ This profile is appropriate for NCD research, as 86% of premature NCD deaths occur between the ages of 30 and 70 years in low-income and middle-income countries³⁷ (table 2).

Table 2. Distribution of participants in the NutriNet-Brasil Cohort enrolled between 26 January 2020 and 12 June 2022, according to sociodemographic variables (n=54 057), compared with the Brazilian population.

Variables	NutriNet-Brasil sample	Brazilian population*
	% (95% CI)	%
Country’s region
Central-West	7.8 (7.6 to 8.0)	7.3
North	2.9 (2.8 to 3.1)	8.3
Northeast	11.2 (10.9 to 11.5)	17.3
South	15.9 (15.6 to 16.2)	14.3
Southeast	62.2 (61.8 to 62.6)	42.1
Sex
Male	20.1 (19.8 to 20.5)	49.0
Female	79.9 (79.5 to 80.2)	51.0
Age group (years)
18–29	22.4 (22.1 to 22.8)	30.4
30–39	28.4 (28 to 28.8)	22.0
40–49	22.8 (22.5 to 23.2)	18.4
50–59	17.7 (17.4 to 18)	13.9
60 or more	8.7 (8.4 to 8.9)	15.3
Level of education†
No education or incomplete elementary school	0.9 (0.8 to 0.9)	46.1
Complete elementary to incomplete higher education	31.3 (30.9 to 31.6)	43.4
Graduated	67.9 (67.5 to 68.3)	10.6

^*Data from the 2010 Brazilian Census available at: https://sidra.ibge.gov.br/

^†Data from the 2010 Brazilian Census available at: https://sidra.ibge.gov.br/

Data from NutriNet-Brasil have already yielded important insights. Two longitudinal studies examined dietary changes and weight gain during the COVID-19 pandemic. The first study observed an increase in the consumption of healthy eating markers (unprocessed and minimally processed foods) and stability in the consumption of unhealthy food markers (UPFs), with differences across regions and levels of education. A trend towards increasing consumption of healthy and unhealthy food markers in the Northeast and North macro-regions and among less educated participants suggested social inequalities in the response to the pandemic.³⁸ The second study showed that 19.7% of the participants gained ≥2 kg during the first 6 months of the pandemic. Weight gain was directly associated with men, lower education and previous overweight status and inversely associated with age.³⁹

Other longitudinal analyses using data from NutriNet-Brasil aimed to assess the prospective association of the two previously mentioned diet quality scores based on the Nova food classification with BMI gain (an increase in BMI of ≥5% during a 15-month period).³⁵ Using information from 9551 participants, the study found a direct and linear relationship between quintiles (Q) of the Nova-UPFs and the risk of BMI gain (relative risk Q5/Q1=1.34; 95% CI 1.15 to 1.56) and an inverse and linear association between quintiles of the Nova-WPFs and the risk of BMI gain (relative risk Q5/Q1=0.80; 95% CI 0.69 to 0.94). A moderate inverse correlation was also observed between the two scores (−0.33). Mediation analysis indicated that 15.0% of the total effect of the Nova-UPFs on the risk of BMI gain could be explained by the Nova-WPFs, and the Nova-UPFs could explain 25.7% of the total effect of the Nova-WPFs.³⁵

Finally, the study by Werneck et al, ³⁴ explored the prospective association between adherence to the ultra-processed dietary pattern and risk of depressive symptoms, using data from 15 960 NutriNet-Brasil participants. The percentage of energy intake from UPF was assessed using the Nova24h, while new cases of depressive symptoms were assessed using the Patient Health Questionnaire-9. In this study, each 10% increase in the dietary share of UPFs was associated with a 10% increased risk of incident cases of depressive symptoms, indicating that an ultra-processed dietary pattern was linked to a higher risk of developing depressive outcomes.³⁴

New studies are planned to further investigate the relationship between diet and a range of health outcomes collected in the cohort according to their incidence. Current analyses have prioritised outcomes with higher incidence rates, such as hypertension, diabetes, obesity, hypercholesterolaemia and gastrointestinal disorders. Other outcomes, including cancer and cardiovascular diseases, will be analysed as the cohort accrues enough cases with longer follow-up. In terms of exposure, we plan to explore not only UPF intake but also other diet quality indicators, including consumption of animal-sourced foods and whole plant-based foods.

Strengths and limitations

The prospective design and large sample size of NutriNet-Brasil are the main strengths of this study. Additionally, the national pioneering in developing a cohort study completely online is noteworthy, as it allows for the quick and convenient collection of data, with reduced costs, and without relying on face-to-face interviews (eg, data collection was not compromised during the COVID-19 pandemic). Finally, new self-reported tools were developed in NutriNet-Brasil, including the Nova24h and the Nova24hScreener, to measure food consumption, considering the degree of food processing and the Nova classification system. Both tools are freely available for researchers’ use in Brazil (in Portuguese only; available at: https://questnova.com.br/⁴⁰

The NutriNet-Brasil digital infrastructure and the large pool of participants have allowed the development of smaller substudies with diverse objectives, including public policy evaluation in Brazil. Such potential was also explored through many cross-sectional, experimental and methodological studies in the NutriNet-Santé cohort, using either subsamples or the entire cohort.⁴¹,⁴⁵ So far, in NutriNet-Brasil, a cross-sectional study with a subsample of 2052 participants was conducted to evaluate the association between adherence to food practices recommended by the Dietary Guidelines for the Brazilian Population and diet quality.⁴⁶ The tool used to measure adherence to the guidelines was also validated in a subsample of the cohort.⁴⁷ Additionally, future studies will evaluate the impact of the new front-of-pack labelling implemented in 2020 in Brazil, as well as the relationship between food consumption at different processing levels and symptoms of food addiction. Finally, a nested randomised controlled trial in the NutriNet-Brasil aiming to evaluate the effect of a digital intervention on the reduction of UPF intake is being devised.

It is important to mention some challenges encountered during the process. Notably, a considerable number of dropouts were observed in the third month after registration, possibly explained by the impulsive registration of some participants without full knowledge of how the research would unfold, influenced by extensive television coverage shortly after its initiation. On careful examination, it became apparent that more detailed and lengthy questionnaires prompted increases in dropout rates. Furthermore, participants were more likely to respond to a subsequent questionnaire if they had completed the preceding one (data not presented). These findings highlight the need to balance questionnaire length and participant burden to minimise dropout.

Regarding the tools that were found to be valid to evaluate food consumption, it is noteworthy that they were developed specifically (but not exclusively) for the NutriNet-Brasil study to capture accurate information on food consumption, particularly on the level of industrial processing of foods. Yet, studies using dietary exposures other than the dietary share of the four Nova groups are also possible and desirable.

There are some limitations to the NutriNet-Brasil cohort that should be mentioned. The study included volunteer participants with Internet access, who were more likely to be women, have higher educational status and live in the Southeast and South regions³⁵ and may, thus, be more health-aware than the general population in Brazil. Data from the Continuous National Household Sample Survey (PNAD) showed an increase in Brazilian households’ Internet access, but disparities related to age group, socioeconomic level and place of residence in the use of this technology persist.²¹ Participant recruitment has been conducted mainly through social media, mass media outlets (TV and newspapers) and posters displayed in primary healthcare settings—a strategy specifically implemented to improve outreach to more marginalised populations (those with lower income, lower educational attainment and residents of remote areas). Despite this effort, it is acknowledged that the use of posters alone was not sufficient to achieve a representative sample of the Brazilian population. On the other hand, individuals without internet access (via cell phone, computer or tablet) were unable to participate in the study. Therefore, estimates of prevalence or incidence obtained from the NutriNet-Brasil cohort, for example, would not be generalisable to the overall Brazilian population. However, the lack of representativeness would be less likely to affect the relationship between dietary patterns and different health outcomes, probably attenuating the magnitude of those associations.⁴⁸ Self-reported information on diet and health could also be a source of bias, but previous validation studies have suggested good agreement between web-based and objectively measured questionnaires.⁴⁹,⁵¹ The confirmation of new health outcome diagnoses and the use of validated tools to measure dietary intake enhanced data quality. Although we did not perform a validation study for the self-report of anthropometric measures, a validation study conducted by the NutriNet-Santé team found high agreement between self-reported weight and height and values measured by a trained professional.⁵² Self-reported measurements of weight and height were shown to be valid alternatives for determining weight status in the Brazilian population, according to data from the Brazilian National Health Survey in 2013.⁵³ In addition, we carried out an agreement study for the usability of our online platform compared with self-reported weight and height through telephone calls. We found an intraclass correlation coefficient of 0.99 for weight and 0.98 for height among NutriNet-Brasil participants (results not published).

In conclusion, the NutriNet-Brasil Cohort Study is a groundbreaking initiative to investigate the effect of dietary patterns on NCDs in Brazil. As the largest web-based study of its kind in the country, NutriNet-Brasil leverages the power of the Internet to enrol over 88 000 participants, offering a unique opportunity to explore the evolving nutritional paradigm and changes in food processing within the Brazilian population. The study’s methodology outlined, in this paper, encompasses innovative recruitment strategies, dynamic data collection through self-administered questionnaires and rigorous data quality control measures. By focusing on dietary patterns and considering the cultural context, NutriNet-Brasil data have the potential to fill notable gaps in existing research and contribute valuable information for health policies and actions. With its commitment to open collaboration, NutriNet-Brasil acts as a significant player in advancing the understanding of the complex interplay between diet and health. As the cohort progresses, the insights gained from NutriNet-Brasil can be a powerful tool to inform evidence-based interventions and shape future health strategies worldwide.

Collaboration

The NutriNet-Brasil team provides opportunities for collaboration with other researchers through proposals that can be sent by email (nutrinetbrasil@fsp.usp.br) and have been evaluated by an internal committee to allow access to the data. The novel tools developed to evaluate food consumption, which are applied in the cohort, are freely available (in Portuguese only) on a specific website (https://questnova.com.br) and on the Nupens/USP website. Other researchers are able to collect and analyse data using both the Nova24h and the Nova24hScreener.

Data availability statement

Data are available upon reasonable request.