Food processing and cardiometabolic risk factors: a systematic review

ABSTRACT OBJECTIVE To systematically review the evidence for the association between food consumption according to processing and cardiometabolic factors in adults and/or the elderly. METHOD Two independent evaluators analyzed the electronic databases PubMed, Web of Science and Lilacs until December 2018. We used the following terms: (convenience foods OR food processing OR highly-processed OR industrialized foods OR minimally-processed OR prepared foods OR processed foods OR ultra-processed OR ultraprocessed OR ultra processed OR unprocessed) AND (metabolic syndrome OR hypertension OR blood pressure OR diabetes mellitus OR glucose OR glycaemia OR insulin OR cholesterol OR triglycerides OR blood lipids OR overweight OR obesity) AND (adult OR adults OR adulthood OR aged OR elderly OR old). We assessed methodological and evidence qualities, and also extracted information for the qualitative synthesis from the selected studies. RESULTS Of the 6,423 studies identified after removing duplicates, eleven met the eligibility criteria. The main food classification we used was Nova. The consumption of ultra-processed foods was positively associated with overweight and obesity, high blood pressure and metabolic syndrome. All articles included met more than 50% of the methodological quality criteria. The quality of evidence was considered moderate for the outcome overweight and obesity and weak for hypertension and metabolic syndrome. CONCLUSIONS The Nova food classification stands out in the area of nutritional epidemiology when assessing the effects of food processing on health outcomes. Although caution is required in the interpretation, the results indicated that the consumption of ultra-processed foods can have an unfavorable impact in the health of individuals.


INTRODUCTION
Cardiovascular diseases (CVD) comprise the main cause of mortality in the world and approximately three quarters of deaths occur in low and middle income countries 1,2 . Risk factors for CVD include behavioral factors, such as unhealthy eating, smoking, physical inactivity and alcohol abuse 1,3 . As a consequence of behavioral risk, the most frequent cardiometabolic factors are high blood pressure (hypertension), dyslipidemia, hyperglycemia, overweight and obesity 1,3 .
Adequate and healthy feeding for a given population involves biological, environmental, social, demographic and economic aspects [4][5][6] . There are changes in eating habits worldwide, characterized by the dominance of products from the food industry [6][7][8] which are not part of the traditional food classification systems 9 . These traditional classifications are restricted to the biological properties of food, i.e., they group food according to the nutrients present in it 9 .
The monitoring of food consumption contributes to the diagnosis of the food and nutritional situation of populations and provides subsidies for the planning and organization of health services and formulation of policies and actions in the field of public health 10 . There are food classifications based on processing 8,11,12 , but there is no synthesis of evidence on the association of food consumption assessed from these classifications with the risk factors for CVD, which are a group of diseases of extreme worldwide relevance 1,3 . Thus, we intended to help elucidate the importance of such classifications in the context of nutritional epidemiology and public health and, in this study, we aimed at conducting a systematic review to assess the association between food consumption according to processing and cardiometabolic factors in adults and/or the elderly.

METHOD
The report of this systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyzes Protocol (Prisma) 13 . The study protocol was submitted to the International Prospective Register of Systematic Reviews (Prospero), being approved under number CRD42019119765.

Search Strategy
We examined the electronic databases PubMed, Web of Science and Lilacs until December 2018. The aim was to conduct a systematic investigation of original studies that assessed the association between food consumption according to its processing and cardiometabolic factors.
To define the search terms, in addition to the Health Sciences Descriptors (DeCS) and Medical Subject Headings (MeSH), we carried out an exploratory investigation aimed at identifying keywords consistently referred to in articles in the area. Therefore, we used the following terms: (convenience foods OR food processing OR highly-processed OR industrialized foods OR minimally-processed OR prepared foods OR processed foods OR ultra-processed OR ultraprocessed OR ultra processed OR unprocessed) AND (metabolic syndrome OR hypertension OR blood pressure OR diabetes mellitus OR glucose OR glycaemia OR insulin OR cholesterol OR triglycerides OR blood lipids OR overweight OR obesity) AND (adult OR adults OR adulthood OR aged OR elderly OR old). Table 1 exemplifies the search strategy in the electronic databases. As additional research, we considered the bibliographic references of the selected articles.

Eligibility Criteria
The eligible studies should present the following characteristics: I) be an original article; II) be conducted in humans; III) address the assessment of the association between food consumption according to processing (exposure) and cardiometabolic risk factors (outcome); IV) present as target population adults and/or the elderly; and V) be published in Portuguese, English or Spanish. The operationalization of the exposure allowed the inclusion of articles that used such a food classification system that considered industrial processing to define groups of foodstuffs. We excluded articles that indirectly assessed consumption, used data on availability, acquisition or commercialization of food, in addition to those whose target population comprised pregnant women or individuals with some special health condition.

Study Selection
Two independent evaluators selected the articles to be included. Faced with cases of disagreement, a third reviewer conducted the trial. Initially, the publications were imported into the EndNote ® version X7 program, in which duplicates were checked, followed by reading of the titles and abstracts. The studies selected in the previous stages according to the eligibility criteria were read in full.

Methodological Quality Assessment
We assessed the methodological quality of the selected articles according to the Strengthening the Reporting of Observational Studies in Epidemiology (Strobe) 14 initiative, which comprises a checklist for observational studies. The maximum score to be obtained is 22 points, distributed as follows: title and/or summary (one item), introduction (two items), methodological aspects (nine items), results (five items), discussion (four items) and other information (one item -on financing) 14 . Each of the 22 items received a score of 0 or 1 considering whether it "does not meet" or "meets" each criterion, respectively. Based on the sum of the checklist's score we established three categories for quality assessment: A, for studies that met more than 80% of the criteria; B, for studies that achieved 50 to 80% of the criteria were considered; and C, for those that met less than 50% of the criteria 15 .

Evidence Quality Assessment
We conducted the quality of evidence for the relationship between consumption of ultraprocessed foods (UPF) and each outcome by using the Grading of Recommendations Assessment, Development and Evaluation (Grade) system 16,17 . Thus, the classification of the studies was carried out as follows: A) high evidence; B) moderate evidence; C) low evidence; and D) very low evidence 16,17 . Observational studies start with low quality of evidence (C) and, among the factors that increase the level of classification, are included the magnitude of the effect, dose-response gradient and plausible confounders that may reduce the demonstrated effect or increase an unobserved effect. Among the aspects that may decrease the level of evidence are methodological quality (risk of bias), inconsistency of results, indirect evidence, imprecision and publication bias 16,17 .

Data analysis
In order to carry out the narrative synthesis of the characteristics as the main results and descriptive approach we extracted the following information from each selected article: name of the main author, year of publication and research data collection, country of study, design, sample size and characteristics, method used to measure exposure, adjustment variables in the analysis and main results.

RESULTS
The search strategy identified a total of 7,216 publications in the electronic databases PubMed, Web of Science and Lilacs. Out of this total, we excluded 793 as they were duplicates, resulting in 6,423 references. After analyzing the titles and abstracts, reading them in full and applying the eligibility criteria, we selected 11 studies. There was no inclusion of articles through the additional search in the reference list of selected articles. We show the complete flowchart of the selection process in Figure.

Characteristics and Methodological Quality of Included Articles
We described the characteristics and methodological quality of the publications included in this review in Table 2. Most studies (six) [18][19][20][21][22][23] were conducted in high-income countries, while five publications [24][25][26][27][28] were held in developing countries contributed, three of which carried out in Brazil 24,25,27 . As for the design, more than 80% were cross-sectional. The two cohort studies identified were conducted with the same population, which consisted of Reason for exclusion: • Food patterns defined a posteriori (9) and a priori (4) not related to processing. • Exposure assessment based on data regarding food availability, purchase or sale (4). • Failure to assess the outcomes of interest (15). • Review article, comment or event summary (3). • Assessment of overweight/obesity as exposure and food consumption according to processing as an outcome (3). • Pregnant women were the target population (2).   According to the Strobe criteria, it is possible to consider that the methodological quality of the analyzed articles was satisfactory, with an overall average of 17.5 points (minimum score: 13 24 ; maximum score: 21 27 ) and none were classified with quality C. Six studies 18,19,21,23,26,27 were considered of quality A and five 20,22,24,25,28 of quality B ( Table 2).
The cardiometabolic risk factors identified were overweight or obesity (eight articles 18,19,22-25,27,28 ) and arterial hypertension (one article 21 ). We included two studies 20,26 that  Results stratified by age. c Cross-sectional analysis in a cohort study. d Self-report of weight and height. e Self-report of weight and height for approximately 37% of the sample.  metabolic syndrome and its components because this condition is characterized by the simultaneous presence of outcomes of interest in this review 29 . Table 3 shows, in chronological order, according to the date of publication of each article and for all dependent variables, the following characteristics of the units of analysis: definition of exposure, adjustment variables in the analysis and synthesis of the main results.
We identified four different methods for assessing exposure in this work, the most frequent being the food frequency questionnaire (FFQ) 21,22,24,26,27 and the 24-hour recall 19,20,23,28 , used in five and four publications, respectively. Research was also identified that evaluated exposure through the food diary 18 and food record 25 . The Nova classification was present in nine studies to define food according to processing [18][19][20][21][22][23][24][25]27 . Another article was based on Nova to determine the food groups, then carrying out an exploratory factor analysis, which identified two dietary patterns: "minimally processed/processed food pattern" and "ultra-processed food pattern" 26 . Finally, one publication defined packaged, frozen, canned, bagged and/or packed foods as processed 28 . We highlight that the association between UPF consumption and the identified health outcomes were assessed by all surveys that used the Nova classification, representing ten of the eleven reviewed surveys.
Although there was some homogeneity in food classification according to the extent and purpose of processing, the operationalization of the exposure was different among the articles analyzed. The main form of assessment was the percentage of total energy intake of the interest groups in each study, which was analyzed on a continuous basis 18,23,28 and/or categorized into quartiles or quintiles 19,20,23,25,27 . There were articles that considered the number of daily servings 21,22 , the consumption greater than or equal to three times a day 24 or based on Nova system to identify eating patterns 26 a posteriori.
Regarding the possible confounders of the association of interest, only one article did not adjust for socioeconomic and demographic factors, as well as for any behavioral variables of cardiovascular risk, such as physical activity, smoking and/or alcohol consumption 24 .
Another five studies 20,22,25,26,28 , although they controlled for socioeconomic, demographic and behavioral confounders, provided estimates only after adjustment for possible mediators, such as energy intake and consumption of other food groups.

Overweight or Obesity
Eight publications investigated food consumption according to processing and overweight or obesity 18,19,[22][23][24][25]27,28 . Most studies included more than one way of defining the outcome, but the adiposity measures considered only anthropometric indicators, evaluating the body  29 was not included in this assessment because exposure was processed foods. No study showed a negative association with the outcomes of interest.
Seven studies found a positive association among the consumption of UPF with, at least, one of the different methodologies for the operationalization of BMI 19,[22][23][24][25]27,28 and abdominal obesity 19,27 . In addition, four articles reported a dose-response gradient for this association 19,22,23,27 , i.e., the higher the consumption category of UPF, the higher the BMI averages 19,27 and WC 19,27 and the higher the risk of overweight 27 , obesity 19,23,27 , overweight 19,22 or abdominal obesity 19,27 . Only one study did not observe a statistically significant relationship between UPF consumption and adiposity measures 18 . However, the higher consumption of processed culinary ingredients (PCI) or the combination of them with minimally processed foods (MPF) provided protection for the evaluated outcome 18 . Finally, the consumption of UPF was associated with a high BMI only among participants in the age groups from 40 to 59 years and above or equal to 60 years in a representative sample of the Brazilian population 25 .

Arterial hypertension
Only one article evaluated arterial hypertension as the main outcome, and it was observed that the higher consumption of UPF (3rd tertile of consumption compared to 1st tertile) increased its incidence (hazard ratio [HR] = 1.21; 95%CI 1.06 -1.37) 21 (Table 3).

Metabolic Syndrome and Components
Two studies evaluated the association between the exposure of interest in this review and metabolic syndrome 20,26 . Table 3 presents an article in which the target population included Canadian Indians that presented positive association between UPF consumption and the outcome (odds ratio [OR] = 1.90; 95%CI 1.14-3.17 ) 20 . However, another study with Lebanese adults did not find a significant relationship between the "ultra-processed foods pattern" and the metabolic syndrome (OR = 1.11; 95%CI 0.26-4.65) 27 . The highest consumption of the "minimally processed/processed pattern" was a protective factor (OR = 0.18; 95%CI 0.04-0.77) 26 for this outcome. In analyzes that considered the presence of each component of the metabolic syndrome, it was found that the consumption of UPF increases the risk of low HDL cholesterol 20 , while consumption of the "minimally processed/processed foods pattern" reduces the chance for low HDL cholesterol and hyperglycemia 26 .
In addition to the heterogeneity of the target population and the definition of exposure, there was also a difference in sample size and methods for collecting consumption data, which were obtained through the 24-hour recall 20 and the FFQ 26 , respectively. The same reference was used in both studies to define metabolic syndrome 29 . Table 4 shows the analysis of the quality of evidence according to the Grade 16,17 system. All included studies were observational and started from a low level of evidence (C). As this is a systematic review article, the evidence is indirect, and the possibility of publication bias is not ruled out; large effect magnitudes have not been identified; considerable methodological limitations were not verified in the reviewed articles, which, for the most part, were adjusted for plausible confounding factors; and the sample sizes allowed precision of the results. For the positive association between UPF consumption and overweight or obesity, a dose-response gradient with consistent results was reported in four studies; thus, the classification was raised to a moderate level of evidence (B). For the relationship between UPF consumption and the outcomes of hypertension and metabolic syndrome, the level of evidence remained low (C).

DISCUSSION
This systematic review identified and summarized the results of 11 studies that assessed the association between food consumption according to processing and cardiometabolic factors in adults and/or the elderly. Three outcomes were verified: overweight or obesity, arterial hypertension and metabolic syndrome. According to the reviewed articles, the level of evidence was considered moderate for the first outcome and low for the other evaluated morbidities.
This is our first systematic knowledge review in which we proposed to assess food consumption according to cardiometabolic processing and outcomes, considering the food classifications defined initially. It is important to highlight that this review did not restrict the search for a food classification. However, especially regarding the UPF group, Nova was used in ten out of the eleven studies included, possibly due to its international recognition and validity in the field of public health and nutritional epidemiology 8,9,30,31 . Nova considers nutritional and non-nutritional attributes of foods that can influence eating behavior, nutritional quality of food and health outcomes 19 . We emphasize that Brazil was a pioneer in using Nova to support national guidelines for food and nutrition 4 .
The unfavorable nutritional profile related to the consumption of UPF, which has an impact on the nutritional quality of food 32 , possibly stimulates the execution of research that evaluates the repercussion of this consumption on negative health outcomes. Thus, the inference of biological plausibility is feasible, considering that the high intake of UPF characterizes a diet with higher concentrations of sodium, sugar, total and saturated fats, with reduced fiber and protein content, highly energetic 18,33-37 , presenting high glycemic index 38 that promotes inflammatory processes resulting from changes in the composition and metabolism of the intestinal microbiota, favoring metabolic disorders 39 . Two other articles found protective effect on the consumption of minimally processed and processed foods by people who are overweight 18 , and those presenting metabolic syndrome and some of its components (hyperglycemia and low HDL cholesterol) 26 .
The available evidence for overweight or obesity was much more abundant in relation to the other dependent variables as identified in eight studies 19,[22][23][24][25]27,28 . The only research that did not report positive association between UPF consumption and the referred outcome analyzed together two groups from Nova (processed and ultra-processed foods) 18 , which may have contributed to such result. After the period stipulated for inclusion of articles in this review, the results of a randomized clinical trial with a crossover methodology were released, which verified an increase in body weight and energy intake of the participants during the two weeks they maintained the diet with consumption of UPF 40 . Thus, the positive association reported by this review with moderate level of evidence corroborates the results of a recent clinical trial 40 , a type of design that can raise the level of evidence 16,17 . It is worth mentioning that an association in the same direction was found in two 41,42 of three cross-sectional articles 34,41,42 which were excluded from this review because they analyzed anthropometric measures of adiposity as exposure to the consumption of UPF. Previous studies of narrative review on UPF consumption and obesity 43 and systematic review on UPF consumption and body adiposity during childhood and adolescence 44 reinforce the results that foods belonging to this group can contribute to increase body adiposity.
For arterial hypertension and metabolic syndrome, one 21 and two 20,26 articles respectively, were identified. The three surveys pointed out that the consumption of UPF increased the risk for the evaluated outcomes 20,21,26 . The level of evidence was low, given that there are not enough studies to guarantee confidence in the results. In one of the articles 26 presenting metabolic syndrome as outcome, the exposure was "ultra-processed foods pattern" and, although Nova was used to define the food groups, exploratory factor analysis added foods that do not belong to the UPF in this pattern, which led to the conclusion that the effect may have been diluted, with a confidence interval that included the unit 26 .
Due to the heterogeneity in the operationalization of the exposure and outcome variables, it was impossible to perform a quantitative synthesis (meta-analysis). This fact introduces a limitation in summarizing these results. As a limitation of this review, we highlight that all articles included presented observational design. We also highlight that experimental studies are justified in the face of considerable observational evidence 45 , and the scarcity of this type of design can be explained by the recent publication dates of the observational studies we identified. In addition, the logistical difficulties of the time between exposure and the incidence of outcomes, added to the ethical reasons that contribute to limit the amount of experimental references that assess the impact of food on health in humans 45 need to be considered. The greater difficulty in publishing studies with negative results or without association and the language restriction in the selection of articles 17 could lead to publication bias, since the search was carried out in indexed electronic databases. However, there is no knowledge of negative results among Brazilian researchers who develop their work in this area of knowledge.
The studies included are mostly cross-sectional, which does not allow attributing causality to the results found, and the limitations inherent to nutritional epidemiology still need to be considered, given the complexity of human nutrition and the difficulty of knowing exactly the real food consumption of individuals 45 . In order to ensure greater consistency of results, only surveys that assessed food consumption at the individual level were eligible. Finally, we emphasize that the main limitation reported in the studies included in this review was related to the collection of exposure data, as the instruments were not designed to obtain consumption information according to the extent and purpose of food processing.
Considering that this is a non-differential classification error, it is possible that the results of the studies have underestimated the magnitudes of association.
As for the five studies 20,22,25,26,28 that presented only controlled estimates for possible mediators and considering the approaches used, this adjustment may lead to an underestimation in the association measure or introduce collision bias in the presence of confounders between the measurer and the outcome46. The association between UPF consumption and cardiometabolic factors should not differ between populations in the biological sense. However, it should be noted that socioeconomic and behavioral variables are important confounding factors. Regarding to the extrapolation of results, there was information from middle and upper income countries for the outcomes of overweight or obesity, from studies with different epidemiological designs, which showed consistency in the different populations evaluated. Such facts allow greater confidence in the generalization of the results for countries with diverse socioeconomic and behavioral characteristics, in addition to reducing the possibility that this association is due to chance or residual confusion. For the other outcomes (hypertension and metabolic syndrome), generalization should be prudent. In addition, we highlight that our proposal was to review articles whose target population consisted of adults and/or the elderly. Among the articles that included both age groups, only two 25,28 reported the number of participants in each group. Thus, we believe that most of the data comes from adult individuals, suggesting that the conclusions would be more appropriate for this age group.
In order to raise the level of evidence and guarantee the temporality and consistency of the results in different confounding scenarios, we recommend the conduction of other studies that, while maintaining the satisfactory methodological quality identified in the articles included in this review, would present longitudinal designs. We also suggest other research designs investigating the consequences of exposure to other Nova food groups.
This review allow us to conclude that the Nova food classification stands out in the area of nutritional epidemiology, which has evaluated the role of food processing and health outcomes, with UPF being more widely studied in relation to other food groups that integrate the classification. The results presented in this review allow us to suppose that the consumption of UPF can have an unfavorable impact on the health of individuals, especially contributing to increase the BMI. Considering the knowledge that diet entails a cardiovascular risk factor that can be modified and that the outcomes assessed in the reviewed studies comprised cardiometabolic factors 1 , in addition to the already described impact described of UPF on cardiovascular disease mortality in the United Kingdom 47 and Brazil 48 , this study may contribute to strengthening scientific evidence that underlies public policies related to the area of food and nutrition and the coping with cardiovascular diseases. In order to reduce the population's consumption of UPF, Brazil has shown important advances 4 , but there are still several challenges to be achieved nationally and internationally 49 .