Consumption of ultra-processed foods and eight-year risk of death from all causes and noncommunicable diseases in the ELSA-Brasil cohort

Abstract Increased consumption of ultra-processed foods (UPF) is associated with higher incidences of many noncommunicable diseases (NCDs) and death from all causes. However, the association between UPF and cardiovascular disease (CVD) mortality remains controversial. Our study investigated whether UPF consumption is associated with a higher risk of death from all causes, NCDs, and CVD. This study includes 14,747 participants from the ELSA-Brasil cohort followed up over an eight-year period. The NOVA classification was used to estimate the proportion of UPF (grams/day) in one’s diet. Cox regression was also applied. After adjustment for sociodemographic, health, and behavioural factors, a 10% increase in UPF in participants’ diets raised the risk of death from all causes and NCDs by 10% (95%CI: 1.01-1.19) and 11% (95%CI:1.02-1.21), respectively. However, UPF consumption was not associated with CVD mortality. The findings support public policies aimed at reducing UPF consumption in an attempt to reduce the NCD burden.


Introduction
Noncommunicable diseases (NCDs) are the leading cause of death in Brazil and worldwide, accounting for 75% of premature deaths (30 to 69 years) in adults (WHO 2018).In 2019, 738,371 deaths from NCDs were reported in Brazil, of which 41.8% occurred prematurely (Ministry of Health of Brazil 2021).Cardiovascular disease (CVD) ranks first among NCDs, negatively impacting the quality of life, productivity, and health system (Mansur et al. 2022).
An inadequate diet is among the leading causes of disability-adjusted life years (DALYs) and deaths from NCDs in Brazil and worldwide (Machado et al. 2022).Changes in eating patterns resulting from an increased consumption of ultra-processed foods (UPF) have been associated with poorer diet quality (Monteiro et al. 2019a;Machado et al. 2022), although the contribution of UPF to the overall diet remains lower in Brazil and other low-and middle-income countries as compared to high income-countries (Monteiro et al. 2019a).UPFs are industrial formulations made mostly or entirely from substances extracted from food or derived from food constituents with numerous additives (Monteiro et al. 2019b).In addition to the poorer nutritional composition (Louzada et al. 2018), evidence suggests that changes resulting from food processing, such as changes in the food matrix and the presence of neoformed contaminants and additives, may contribute to the deleterious effect of these foods on health (Fardet et al. 2015).
Systematic review studies (Chen et al. 2020;Silva Meneguelli et al. 2020) and recent meta-analyses (Lane et al. 2021;Pagliai et al. 2021) have shown that the consumption of UPF is associated with numerous NCDs, such as being overweight and obesity, hypertension, diabetes, metabolic syndrome, CVD, depression, and cancer.The consumption of these foods also appears to increase the risk of death from all causes (Blanco-Rojo et al. 2019;Kim et al. 2019;Rico-Campà et al. 2019;Schnabel et al. 2019;Bonaccio et al. 2021;Romero Ferreiro et al. 2021) and CVD (Bonaccio et al. 2021;Zhong et al. 2021;Juul et al. 2021a).
Despite the evidence linking UPF consumption to various NCDs and overall mortality, no studies have been found that have investigated the association between the consumption of these foods and the risk of death from NCDs.Furthermore, studies that have investigated the association of UPF consumption with the risk of death from CVD (Kim et al. 2019;Rico-Campà et al. 2019;Bonaccio et al. 2021;Zhong et al. 2021;Juul et al. 2021a) presented conflicting results.While the Framingham Offspring Cohort (Juul et al. 2021a) and two other longitudinal studies (Bonaccio et al. 2021;Zhong et al. 2021) suggest that higher UPF consumption is associated with a higher risk of death from CVD, the results of the National Health and Nutrition Examination Survey (NHANES III) (Kim et al. 2019) and the Seguimiento University of Navarra (SUN) (Rico-Campà et al. 2019) indicated no association between UPF consumption and CVD mortality.
Thus, the present study sought to investigate whether UPF consumption at baseline (first visit) from the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil) was associated with the risk of death from all causes, NCDs, and CVD in approximately eight years of follow-up.The hypotheses are that higher UPF consumption would be associated with higher risks of death from all causes, NCDs, and CVD.

Study population
This prospective study includes 15,105 participants from the first visit (2008)(2009)(2010) of the ELSA-Brasil cohort.Participants answered structured face-to-face interviews and underwent a series of clinical and laboratory examinations.All measurements were taken by a trained and certified research team.Detailed information on the ELSA-Brasil cohort is available in previous publications (Aquino et al. 2012;Schmidt et al. 2015).ELSA-Brasil was approved by the Research Ethics Committees of the participating teaching and research institutions (USP: 669/06; ENSP/Fiocruz: 343/06; UFMG: 186/06; UFES: 041/06; UFRGS: 06-194; UFBA: 027-06).This study was performed in line with the principles of the Declaration of Helsinki, and all participants signed an Informed Consent Form.
To properly analyse the outcomes of interest, deaths from all causes, NCDs, and CVD, all participants during the first study visit were eligible.The following were excluded from the study: missing information on food consumption (n = 36); caloric intake less than or equal to the first percentile (n = 150), or greater than or equal to the 99th percentile (n = 150) of total diet kcal sex-distribution; and participants with less than one-year of follow-up (n = 22).Considering some overlap, the final sample included 14,747 participants (Supplementary Figure 1, online supplementary material).The number of missing observations in the covariates of this study was minimal, accounting for less than 2% in the present study.In the multivariate analysis, participants with missing data in any of the covariates were automatically excluded from the model, and the final numbers of participants included in the final models are displayed in the corresponding tables.

Outcome ascertainment
The response variables for this study were the time until death from all causes, NCDs, and CVD.Mortality data were obtained up to December 31, 2018, via annual follow-up calls and hospital record investigations (Barreto et al. 2013).The underlying cause of death was defined according to the International Classification of Diseases, tenth revision (ICD-10), based on a review of death certificates and hospital records or by cross-referencing with data from the Ministry of Health's Mortality Information System.Deaths from NCDs -which include deaths from cancer; blood disorders, and some immune disorders; endocrine and metabolic disorders; mental disorders; CVD; nervous, respiratory, digestive, and genitourinary system disorders; diseases of the musculoskeletal system and connective tissue; and diseases of the skin and subcutaneous tissue -were identified (Supplementary Table 1, online supplementary material).Participants in this study were followed up from the second year of entry into the cohort up to the date of death, date of last follow-up contact, or December 31, 2018, whichever occurred first.

Exposure ascertainment
The explanatory variable was the percentage of UPF contribution, in grams/day (g/d), to the total daily intake of food and beverage.Information on dietary intakes was obtained during the first visit, using a validated semi-quantitative Food Frequency Questionnaire (FFQ) containing 114 food items (Molina, Benseñor, et al. 2013;Molina, Faria, et al. 2013).The FFQ evaluated the participants' customary consumption for the 12 months prior to the interview (Molina, Faria, et al. 2013).
UPF consumption was estimated using NOVA, a classification that distributes foods into four specific groups according to the nature, extension, and purpose of processing (Monteiro et al. 2019b).Information on the ELSA-Brasil FFQ classification according to NOVA is available in a previous publication (Simões et al. 2018;Scaranni et al. 2021) and in the online supplementary material (Supplementary Table 2, online supplementary material).The proportion of UPFs in one's diet was estimated by adding the amount consumed (g/d) of each food item belonging to the UPF group and dividing the sum by the total amount of food and beverage ingested (g/d).In the present study, the consumption in g/d of UPF was adjusted by total caloric intake according to the residual method as proposed by Willett (2012).By definition, the UPF residuals provide a measure of UPF intake uncorrelated with total caloric intake (Willett 2012).

Covariates
All covariates used in the study were obtained during the first visit and included: sex, age (continuous), Brazilian region of the investigation centre (South, Southeast, and Northeast), education (complete college, complete high school, complete elementary school, and incomplete elementary school), free time physical activity measured using the International Physical Activity Questionnaire (IPAQ) (International Physical Activity Questionnaire 2005) (active: ≥150 min/week of moderate activity or ≥75 min/week of vigorous activity and insufficiently active), smoking (never smoked, former smoker, smoker) (Aquino et al. 2012), and excessive alcohol consumption (excessive: men ≥210 g alcohol/week or women ≥140 g alcohol/ week, and not) (Schmidt et al. 2015).The self-reported medical diagnosis of diabetes (no and yes), hypertension (no and yes), hypercholesterolaemia (no and yes), CVD (no and yes), and cancer (no and yes) were included.According to previous literature, all of these variables were considered to be confounders of the investigated associations and included in the final model, regardless of statistical significance.
The variables of body mass index (BMI) (kg/m 2 ) and waist circumference (WC), in their continuous form, were not included in the model, since they are potential mediators of the association between UPF consumption and the outcomes under study.Adjustments for these variables were, however, presented in additional analyses.

Statistical analysis
The characteristics of the study population during the first visit were described using either means and standard deviations (SD) or medians and interquartile ranges (IQR) for continuous variables and proportions for categorical variables.
The association between UPF consumption and the risk of death from all causes, NCDs, and CVD was investigated using the Cox proportional hazards model.UPF consumption was treated as a continuous variable, and its results were expressed as an absolute increment of 10% in the proportion of UPFs in one's diet.The raw coefficients of hazard ratios (HR) and their 95% confidence interval (CI) were estimated, and adjustments were made for confounders.
Restricted cubic spline Cox models with 3 knots, located at the 10th, 50th, and 90th percentiles of UPF consumption, were used to verify a possible non-linear relationship between UPF consumption and the outcomes.The daily UPF consumption of 10% was established as a reference in the restricted cubic spline Cox models, similar to that adopted in other studies (Srour et al. 2019;Bonaccio et al. 2021).Likelihood ratio tests were used to confirm the linearity assumption.
Risk subdistribution models, equivalent to the Fine and Gray models (Fine and Gray 1999), were used to assess the presence of competing risks in the associations between UPF consumption and the risk of death from NCDs and CVD.The original results were maintained, since the risk subdistribution analyses did not indicate the presence of competing risks.The proportionality of risks was assessed using Schoenfeld residuals.The variables of smoking and cancer did not meet the proportional assumptions in the Cox regression models related to deaths from all causes, and smoking and hypertension also failed to meet the proportionality assumptions in the analysis related to NCD death.For these reasons, they were entered as strata in each corresponding model.
An additional analysis was run, including adjustments for BMI and WC, and a sensitivity analysis was also performed, excluding participants with prevalent CVD and cancer during the first visit.The analyses were performed using the statistical program STATA 14.0 (Stata Corporation, College Station, USA) with a significance level of 5%.

Results
The characteristics of the study population are presented in Table 1.Among the 14,747 participants, the mean age was 52.0 (±9.1) years, and the majority were female (54.4%) and had a complete college education (53.0%).The median consumption of UPFs, adjusted for total caloric intake, was 393.9 g/d, corresponding to 18.1% of the total food and beverages consumed daily.The proportion of UPF in the diet was smaller among individuals with low levels of education and who lived in the Northeast region.The main food groups that contributed to UPF consumption among the study participants were: soft drinks and artificial juices (34.5%), ultra-processed bread (16.6%), and ultra-processed dairy products (15.5%) (Table 2).
The median follow-up time was 8.0 (7.9-8.3)years.A total of 510 deaths occurred during the study period, 449 from NCDs and 150 from CVD, with cancer being the most common cause of NCD deaths (46.1%).
The associations between UPF consumption and the risk of death from all causes NCDs, and CVD are presented in Table 3.The linearity assumptions in the relation between UPF consumption and the outcomes were confirmed for deaths from all causes (p-value = 0.31) (Figure 1A), NCDs (p-value = 0.50) (Figure 1B), and CVD (p-value = 0.08) (Figure 1C).Results indicate that every 10% increase in the proportion of UPFs in the diet was associated with a 10% (95% CI: 1.01-1.19)and 11% (95% CI: 1.02-1.21)higher risk of death from all causes and from NCDs, respectively, after adjusting for all confounders.No statistical association was found between UPF consumption and risk of death from CVD during the study period (Table 3).
Additional analyses, including adjustment by BMI or WC, indicated no changes in the results (Supplementary Table 3, online supplementary material).The sensitivity analyses excluding participants with CVD or cancer during the first visit indicated a slight increase in the magnitude of HR for deaths from all causes and NCDs (Supplementary Table 4, online supplementary material).

Discussion
The results from a large multicentric, multiracial cohort of Brazilian adults, who underwent follow up for over eight years, indicate that every 10% increase in the proportion of UPF (g/d) in one's diet increases the risk of death from all causes and from NCDs by approximately 10% after considering a large number of confounders.The cubic spline models indicated linear relationships between UPF consumption and the risk of death from all causes and NCDs.However, no association was observed between UPF consumption and the risk of death from CVD.
The results from the present work are consistent with those from other studies, which investigated the relationship between UPF consumption and the risk of death from all causes (Blanco-Rojo et al. 2019;Kim et al. 2019;Rico-Campà et al. 2019;Schnabel et al. 2019;Bonaccio et al. 2021;Romero Ferreiro et al. 2021).Schnabel et al. (2019), when analysing data from the NutriNet-Santé cohort, found a 14% (95% CI: 1.04-1.27)more significant risk of death from all causes for each 10% increase in the proportion of UPFs (g/d) in one's diet.However, their result lost statistical significance after excluding participants with cancer and CVD.Analysis of the Moli-sani cohort also reported a 26% (95% CI: 1.09-1.46)higher risk of death from all causes associated with a higher proportion of UPFs (g/d) in one's diet (4 th quartile: >14.6% vs. first quartile: <6.6%) (Bonaccio et al. 2021).This latter study found an even higher risk of death from CVD (HR: 1.65; 95% CI: 1.29-2.11)and cerebrovascular diseases (HR: 1.63; 95% CI: 1.19-2.25)associated with higher UPF consumption (Bonaccio et al. 2021).Comparable results were also found in  other studies, such as SUN and NHANES III, which evaluated UPF consumption in different ways, reinforcing the harmful effect of its consumption on the overall risk of death (Kim et al. 2019;Rico-Campà et al. 2019;Romero Ferreiro et al. 2021).Although we did not identify studies that investigated the relationship between UPF consumption and the risk of death from NCDs, several studies have shown the association of UPF with cardiometabolic risk factors and the incidence of some specific NCDs, such as CVD (Srour et al. 2019), diabetes (Srour et al. 2020), and hypertension (Mendonça et al. 2017;Scaranni et al. 2021), indicating the biological plausibility of the association with NCD death observed in this study.One study based on SUN data showed that a higher consumption of UPF (5 servings/day vs. 2.1 servings/day) was associated with a 21% (95% CI: 1.06-1.37)increased risk of hypertension in individuals without comorbidities at the beginning of the study (Mendonça et al. 2017).In ELSA-Brasil, higher UPF consumption (35.0% vs. 14.5% of total calories) was associated with a 23% (95% CI: 1.06-1.44)increased risk of hypertension after four years of follow-up (Scaranni et al. 2021).Additionally, the NutriNet-Santé cohort results showed that a 10% increase in the proportion of UPF (g/d) in one's diet was associated with a 15% (95% CI: 1.06-1.25)higher risk of diabetes (Srour et al. 2020).
However, the results differ among studies investigating UPF consumption's association with CVD mortality.Similar to the present study, the NHANES III (Kim et al. 2019) and SUN (Rico-Campà et al. 2019) studies found no association between UPF consumption and CVD mortality.However, the Framingham Offspring Cohort (Juul et al. 2021a) and other longitudinal studies (Bonaccio et al. 2021;Zhong et al. 2021) reported a positive association between UPF consumption and an increased risk of death from CVD.In the same light, high UPF consumption was associated with a greater incidence of CVD overall and with coronary and cerebrovascular diseases in the NutriNet-Santé cohort (Srour et al. 2019).The inconsistencies observed in these results may be explained by sociodemographic disparities between the study populations, differences in UPF consumption, and the ways of measuring UPF consumption, as well as variations in the confounders included in the analysis.Moreover, the relatively low number of deaths from CVD observed in some studies, including ours, reduced the statistical power to identify associations, especially those of low magnitude (Doll 1998).
Although UPF consumption was not associated with the risk of death from CVD in the present study, models using cubic splines suggest that there may be a non-linear relationship (borderline p-value = 0.08).Zhong et al. (2021) found that consuming 2.4 servings/day and 2.3 servings/day of UPF increased the risk of death from CVD and heart disease, respectively.On the other hand, findings from Srour et al. (2019) and Bonaccio et al. (2021) identified a linear relationship between UPF consumption and the risks of CVD events and mortality.These distinct findings indicate the need to investigate the relationship between UPF consumption and CVD death, especially as there is an upward trend in consuming these foods in many countries.
The main mechanisms concerning the harmful effect of UPF on the risk of death from all causes and NCDs are related to their poor nutritional characteristics, such as high energy density and high concentrations of fat, added sugar, and sodium (Chien et al. 2013;Mozaffarian et al. 2014;Yang et al. 2014;Nagai et al. 2016).In addition, recent evidence suggests that changes in the food matrix caused by food processing, such as in the structure and size of food particles, can interfere in nutrient bioavailability, satiety, and antioxidant capacity, as well as affect the intestinal microbiota, which could promote a pro-inflammatory response, favouring the onset of NCDs (Santos et al. 2020;Lane et al. 2021;Pagliai et al. 2021;Juul et al. 2021b).Cell and animal experiments have also shown that some artificial food additives present in UPFs, such as sodium nitrate and titanium dioxide, have carcinogenic properties.By contrast, others, specifically emulsifiers, appear to be related to the induction of inflammation and cardiometabolic disorders (Lane et al. 2021).Other additives, such as inorganic phosphate salts, have been associated with a higher risk of CVD, kidney disease, and other NCDs (Calvo et al. 2014).Additionally, exposure to exogenous chemicals, such as acrolein and acrylamide, has been associated with a higher risk of cancer (Lane et al. 2021) and CVD (Pagliai et al. 2021).Other compounds commonly present in UPF plastic packaging, such as bisphenol A, have also been associated with a higher risk of cardiometabolic disorders (Rancière et al. 2015).
It is important to note that high UPF consumption contributes to a lower consumption of unprocessed foods (Monteiro et al. 2019a).Fortunately, recent studies have shown benefits related to replacing UPF with other food groups.For example, Blanco-Rojo et al. (2019) demonstrated in hypothetical models that the replacement of UPFs with fresh or minimally processed foods, considering the same caloric supply, resulted in a reduction in the risk of death from all causes among the participants from the Study on Nutrition and Cardiovascular Risk in Spain (ENRICA).Studies considering different scenarios of the home purchase of food in the United Kingdom (Moreira et al. 2015) and Brazil (Moreira et al. 2018) also presented essential estimates of the reduction in the risk of death from CVD associated with a decrease in the consumption of salt, fats, and added sugar found in UPFs.Another study also showed that reducing the caloric intake of UPFs by 10% to 50% could prevent from 5,900 to 29,300 premature deaths in Brazil (Nilson et al. 2023).
Although Brazil has a lower UPF consumption when compared to high-income countries (Monteiro et al. 2019a;Nilson et al. 2023), the relations between the consumption of these foods and deaths from NCDs and all causes are linear, i.e. the risk increases as the proportion of UPF in one's diet increases.Thus, the low consumption of UPF in Brazil represents a window of opportunity for population-based interventions to promote healthy food and prevent the increase in UPF intake (Nilson et al. 2023).For instance, the results of the last household budget survey indicated a deceleration in UPF growth consumption (IBGE 2020), most likely related to public policies implemented in recent years based on the new food guidelines for the Brazilian population (Ministry of Health of Brazil 2014).Nonetheless, current public policies cannot stem the rise in UPF consumption and its negative effect on health (Popkin et al. 2021).

Strengths and limitations
This study stands out for its relatively large sample, for having used a validated instrument to assess food consumption (Molina, Benseñor, et al. 2013), and for having applied the NOVA classification, which is most commonly used in the scientific literature for classifying foods according to the degree of processing (Monteiro et al. 2019a).This study also stands out for having evaluated the association between UPF consumption and the risk of death from NCDs, thus raising the specificity of the association that has already been observed between UPF consumption and the risk of death from all causes.Reverse causality is unlikely, whether by the study's longitudinal design or by the exclusion of subjects with a follow-up time of less than one year.Furthermore, the sensitivity analyses confirmed the results, excluding subjects with cancer or CVD during the first visit.Finally, the association estimates were adjusted for important confounders.
The limitations include the use of FFQ, which needed to be prepared considering the use of NOVA, which may result in a non-differential misclassification and, consequently, an underestimation of the magnitude of the associations (Szklo and Nieto 2012).The use of food consumption data retrieved only from the first visit also hinders the possibility of considering changes that occurred over time, which may underestimate the real impact of UPFs on the risk of death when considering the trend towards a greater consumption of UPFs in the country (IBGE 2020).It is essential to highlight that the data on UPF consumption were obtained through self-reporting, and the instrument used favours the overestimation of general food consumption (Willett 2012).However, it is unlikely that such an overestimation would affect UPFs differently.Therefore, using the UPF consumption percentage in one's diet makes it unlikely that such a limitation could interfere in the results presented in this study.The relatively low number of deaths from CVD (n = 150) may partly explain the lack of association between UPF consumption and the risk of death from CVD.Finally, despite adjustments for essential confounders, a certain degree of residual confounding may exist.
The results of the present study suggest that the increased proportion of UPFs in one's diet is associated with a higher risk of death from all causes and NCDs after eight years of follow-up.These findings reinforce the need to develop and implement more rigorous public policies, such as raising taxes on UPFs, imposing restrictions on UPF advertising, food education for the population, and other actions to support and encourage the consumption of fresh or minimally processed foods.These measures are essential to reducing the global burden of NCDs and the risk of death associated with these diseases.

Figure 1 .
Figure1.associations between energy-adjusted uPf consumption and risk of death from all causes (a), ncds (B), and cVd (c) in the elSa-Brasil.risk estimates were obtained from restricted cubic spline models adjusted for sex, age (years), research centre (by region), schooling, physical activity, smoking, excessive alcohol use, self-report of diabetes, hypertension, hypercholesterolaemia, cVd, and cancer.energy-adjusted uPf consumption was used continuously, per 10% increment of uPf in the diet, with a reference value of 10%. the solid line represents the Hrs, and the dashed lines indicate the 95% cI.three knots were used, located at the 10th, 50th, and 90th percentiles of uPf consumption.Abbreviations: cI: confidence interval; cVd: cardiovascular disease; elSa-Brasil: Brazilian longitudinal Study of adult Health; Hr: hazard ratio; ncds: noncommunicable diseases; uPf: ultra-processed food.
Proportion (g/d) of uPf in the diet considering adjustment for residual method.Abbreviations: cVd: cardiovascular disease; IQr: interquartile range; Sd: standard deviation; uPf: ultra-processed food.

Table 3 .
association between energy-adjusted uPf consumption and risk of death from all causes, ncds, and cVd.elSa-Brasil.