Determinants of dietary behavior among youth: an umbrella review

Background The literature on determinants of dietary behavior among youth is extensive and unwieldy. We conducted an umbrella review or review-of-reviews to present a comprehensive overview of the current knowledge. Methods Therefore, we included systematic reviews identified in four databases (i.e. PubMed, PsycINFO, The Cochrane Library and Web of Science) that summarized determinants of observable child and adolescent dietary behaviors. Data extraction included a judgment of the importance of determinants, strength of evidence and evaluation of the methodological quality of the eligible reviews. Results In total, 17 reviews were considered eligible. Whereas social-cognitive determinants were addressed most intensively towards the end of the 20th century, environmental determinants (particularly social and physical environmental) have been studied most extensively during the past decade, thereby representing a paradigm shift. With regard to environmental determinants, mixed findings were reported. Sedentary behavior and intention were found to be significant determinants of a wide range of dietary behaviors in most reviews with limited suggestive evidence due to the cross-sectional study designs. Other potential determinants such as automaticity, self-regulation and subjective norm have been studied in relatively few studies, but results are promising. Conclusion The multitude of studies conducted on potential determinants of dietary behavior provides quite convincing evidence of the importance of several determinants (i.e. quite some variables were significantly related to dietary behavior). However, because of the often used weak research designs in the studies covered in the available reviews, the evidence for true determinants is suggestive at best.


Background
Dietary behaviors have been found to track from childhood into adulthood [1]. Unhealthy food habits in childhood, therefore, can have a tremendous health impact later in life. Given the high prevalence of nutritionrelated disease and mortality in Western countries [2], it is necessary to develop effective behavioral interventions to improve diet quality. But which factors determine a person's dietary behavior? Interventions to improve health-related behaviors should be tailored to the most important and changeable determinants of these behaviors, preferably applying behavior change theories [3].
To facilitate improvement of relevant, effective programs and policies promoting healthy eating targeting dietary behavior it is important to identify the various factors that may influence children's and adolescents' food consumption.
Socio-cognitive models of (health) behavior and behavior change, such as the Theory of Planned Behavior [4], Social-Cognitive Theory [5], and the Health Belief Model [6] have been applied frequently in development of nutrition education interventions. In very general terms -and not paying attention to the richness of and also differences between these models-these theories regard nutrition behavior to be determined by beliefs and conscious decisions, rational considerations of pros and cons of the behavior, perceived social influences, and assessment of personal efficacy and control. In additional fields of research, the physiological and affective influences on dietary behaviors have been studied, providing evidence for such basic factors as hunger and satiety, sensory perceptions, and perceived palatability of foods [7] as important drivers of food choice and dietary behaviors. And somewhat more recently, the so-called food environment that defines the availability and accessibility (i.e. physical environment), affordability (i.e. economic environment), social appropriateness or support (socialcultural environment), as well as rules, regulations and policies (i.e. political environment) regarding food choice and dietary behaviors has been studied in relation to food intake and dietary behaviors, as informed by (social) ecological behavior models [8][9][10][11][12]. Kremers and colleagues [13] proposed to integrate these insights in their Environmental Research framework for weight Gain prevention (EnRG; Figure 1). EnRG is a dual-process model and regards dietary behavior and physical (in) activity to be the result of direct 'automatic' responses to environmental cues (e.g. meal patterns and routines) as well as of more rational decision making based on cognitions such as intentions and beliefs. Furthermore, EnRG includes mediating pathways between environment and cognitions as well as potential moderators of the impact of these determinants such as habit strength and selfregulation skills.
The purpose of this study was to get a comprehensive and systematic overview of the scientific literature on correlates (referred to as potential determinants) and determinants of dietary behavior among children and adolescent (referred to as youth) to facilitate the improvement of effective healthy eating promoting interventions and identify gaps for future research initiatives. Because the scientific literature on this topic is unwieldy and has been documented in a number of systematic reviews in recent years, we aimed to conduct a review-ofreviews to provide a more comprehensive overview. We were interested in the association of all determinants that are potentially modifiable (social-cognitive, environmental, sensory and automatic processes) with observable dietary behavior (actual consumption behaviors like fruit consumption, beverage intake, snacking) among youth. By conducting a review-of-reviews, the so-called umbrella review, we aimed to (a) explore which determinantbehavior relationships have been studied so far, and (b) assess the importance and strength of evidence of potential determinants. The EnRG framework served to categorize the findings. Parallel to this umbrella review, a separate review-of-reviews of studies among adults was conducted by the same team with the same methodology [14]. Some parts of these two reviews -especially the description of the methodology-are therefore very similar.

Search strategy and eligibility criteria
To identify all relevant systematic reviews, we conducted systematic searches in the bibliographic databases PubMed, PsycINFO (via CSA Illumina), The Cochrane Library (via Wiley) and Web of Science for articles published between January 1, 1990 and May 1, 2014. The search terms included controlled terms, e.g. MeSH in PubMed, Thesaurus in PsycINFO, as well as free text terms (only in The Cochrane Library). Search terms expressing 'food and dietary behavior' were used in combination with search terms comprising 'determinants', 'study design: (systematic) review', 'study population: humans' and 'time span (January 1, 1990 to May 1, 2014)'. The PubMed search strategy can be found in Table 1. The search strategies used in the other databases were based on the PubMed strategy.
Studies were included if they met the following criteria: (i) systematic reviews on observable food and dietary behavior (i.e. consumption behaviors like fruit intake and snacking consumption, not purchasing behavior); (ii) studies describing potential behavioral determinants; (iii) study design: (systematic) review; (iv) study population: humans and (v) time span: January 1, 1990 to May 1, 2014. We excluded: (i) studies that were not written in English; (ii) studies in which dietary behavior was not an outcome of the study; (iii); studies about dietary behaviors in disease management and treatment; (iv) studies that focused on specific population groups (e.g. chronically ill, pregnant women, cancer survivors); (v) studies not published as peer reviewed systematic reviews in scientific journals, e.g. theses, dissertations, book chapters, nonpeer reviewed papers, conference proceedings, reviews of case studies and qualitative studies, design and position papers, umbrella reviews; (vi) reviews of studies on not directly observable dietary behavior (e.g. nutrient or energy intake, appetite); (vii) reviews of studies on nonmodifiable determinants (e.g. physiological, neurological or genetic factors); (viii) reviews of studies on the effect of interventions (but reviews of experimental manipulation of single determinants were included); (ix) reviews not conducted systematically (search strategy including keywords and databases used not identified, and/or with too little information of the included studies presented). The current umbrella review focuses on youth (<18 years). A second umbrella review using the same methodology about determinants of dietary behavior in adults is published elsewhere [14]. Figure 2 summarizes the manuscript selection process. In total, 17714 citations were obtained using PubMed (n = 13156), PsycINFO (n = 961), The Cochrane Library (n = 920), and Web of Science (n = 2677). The subsequent screening of the citations was performed by multiple reviewers (all citations were screened by ES and WK; some were screened by LB, SK, and EV). All titles of the citations were independently screened for relevance by two reviewers (ES and WK). Any disagreement was resolved by including the citation into the abstract screening process. Subsequently, abstracts of the remaining 1031 citations were retrieved for further screening. Another 729 citations were removed, resulting in 292 articles for fulltext assessment for eligibility. In case of doubt, potential inclusion was discussed with a third reviewer (SK). Studies that did not meet the inclusion criteria (n = 257) were removed. Figure 2 displays the reasons for exclusion. Additionally, duplicates (n = 10) were removed. Thereafter, the reference lists of all review papers selected for inclusion (n = 25) were scanned for further relevant references. This reference tracking technique resulted in one additional review article appropriate for inclusion. In total, 26 reviews were considered eligible. However, of these reviews, 9 were only focused on determinants of adult dietary behavior (references reported in the umbrella review about determinants of dietary behavior in adults of Sleddens et al. [14]). Five reviews assessed dietary behavior of both youth and adults [15][16][17][18][19]. Therefore, 17 reviews were considered eligible for our umbrella review on determinants of youth dietary behavior [15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31].

Data extraction including rating of methodological quality
Four authors (ES, WK, LK, and LB) extracted data from the selected reviews. The extracted data included search range applied, total number of studies included in the reviews and number of studies included in the reviews that are eligible for the current umbrella review, total number of participants of included studies in the reviews and number of participants of the included studies that are eligible for the current umbrella review, and age and continent of included eligible studies. For a description of the results, correlate and outcome measures were extracted, as well as overall results of the reviews and overall limitations and recommendations of reviews. Additionally, the methodological quality of the reviews was evaluated using quality criteria adapted from De Vet, De Ridder, & De Wit [32] and based on the Quality Assessment Tool for Reviews [33]. In total, a review was scored on eight criteria (with a total quality scoring ranging from 0-8) (see Table 2); 0 when the criteria was not applicable for the included review; 1 when the criteria was applicable for the included review. Disagreement between the reviewers on individual items were identified and solved during a consensus meeting. The quality of the reviews could be labeled as weak (quality scores ranging from 0 to 3), moderate (quality scores ranging from 4 to 6) or strong (quality scores ranging from 7 to 8). Furthermore, we judged the importance of included determinants in the reviews and judged its strength of evidence. The importance of a determinant refers to the statistical significance of a potential determinant and/or effect size estimate in relation to a particular type of dietary behavior. It refers to the amount of reviews (or eligible studies within the reviews) that did or did not find statistically significant results. For a particular determinant to receive the highest ranking (highest level of importance), all eligible studies in each review should have found a significant relationship and/or reported a (non)-significant effect size larger than 0.30. The strength of evidence represents the consistency between study findings and designs of the studies. Longitudinal observational studies and -where relevant-experimental studies of sufficient size, duration and quality showing consistent effects were given prominence as the highest ranking study designs. For this judgment we applied two coding schemes, see Tables 3 and 4 respectively. The criteria for grading evidence were adapted from those of the World Cancer Research Fund [34]. The data extraction method is similar to the study of Sleddens et al. [14].  Note: 1 A search is rated clearly defined if at least search words and a flow chart is presented; 2 A search is rated as comprehensive if at least two databases and the reference lists of examined papers were searched. *Reviews also including adults; weak (score ranging from 0-3) n = 1 (5.88%), moderate (score ranging from 4-6) n = 9 (52.94%), strong (score ranging from 7-8) n = 7 (41.18%).

Description of reviews
Quality assessment ratings are presented in Table 2.
One review received a quality rating of 2 (weak). The other reviews were rated as moderate (n = 9) or strong (n = 7). In all reviews, the inclusion and exclusion criteria were clearly stated and the review did integrate findings beyond describing or listing findings of primary studies. Clearly defined search strategies were absent in more than half of the reviews (9 out of 17 reviews), as usually a flow chart of the data screening process was missing. Table 5 provides an overview of the characteristics of the included reviews. In three reviews [22][23][24] all included studies were eligible for our current study. Most of the studies included in the reviews used a cross-sectional study design. Six studies did not provide any information about sample sizes [15,16,19,21,24,29]. The remaining reviews included a total sample size of 695 to 570,403. The target groups of the eligible studies ranged in ages between the different reviews, although the focus was on primary school-aged children and adolescents. Most of the studies included in the reviews were conducted in North-America, followed by Europe.  ++ The variable has been found to be a statistically significant determinant in all identified reviews, without exception. This could mean that only one review has included a particular variable, and showed that this was a significant correlate and/or reported a (non)-significant effect size larger than 0.30, but it could also mean that a number of reviews were conducted that included this variable and all of them concluded that the variable was significantly related to the particular behavioral outcome.
+ The variable has been found to be a statistically significant determinant and/or reported a (non)-significant effect size larger than 0.30 in most reviews or studies within the review, with some exceptions. This implies that > 75% of the available reviews concluded the variable to be related, or the separate reviews report that 75% or more of the original studies concluded the factor to be related. This could therefore mean that only one review has included a particular variable, and showed that this was a significant correlate in > 75% of studies. But it could also mean that a number of reviews were executed towards this variable and most, but not all, concluded that the variable was significantly related to the particular behavioral outcome.
0 The variable has been found to be a determinant and/or reported a (non)-significant effect size larger than 0.30 in some reviews (25% to 75% of available reviews or of the studies reviewed in these reviews), but not in others. This could mean that only one review has included a particular variable, and showed 'mixed findings', but it could also mean that results are mixed across reviews.
-The variable has been found not to be a determinant, with some exceptions. This implies that <25% of the available reviews or of the original studies in the included reviews concluded that the variable was related. This could thus mean that only one review has included a particular variable, and generally showed 'null findings', with some exceptions. But it could also mean that a number of reviews were executed towards this variable and most, but not all, concluded that the variable was not significantly related to the particular behavioral outcome.
--The variable has been found not to be related to this particular outcome. The absence of a relation was identified in all identified reviews, without exception. This could mean that only one review has included a particular variable, and showed that this correlate was not related to the behavior in question, but it could also mean that a number of reviews were executed towards this variable and all of them concluded that the variable was unrelated to the particular behavioral outcome. Ideally the definition of the strength of evidence should be based on a relationship that has been established by multiple randomized controlled trials of manipulations of single isolated variables, but this type of evidence is often not available.
The following criteria were used to describe the strength of evidence in this report.    <500 n = 24, 500-1000 n = 20, >1000 n = 53, NR: n = 1/<500 n = 24, 500-1000 n = 20, >1000 n = 53, NR: n = 1 NR North-America n = 50, Europe n = 31, Australasia n = 16, South-America n = 1 Note: Designs of studies: cross-sectional, longitudinal observational, case control, and intervention studies (experimental, behavioral laboratory, filed studies in which interventions were studied); NR: not reported; we were mainly interested to provide a more thorough description on the eligible studies of the included reviews (designs of studies, ages, continent). Table 6 provides an overview of the correlates and outcomes (i.e. observable dietary behaviors) included in the reviews, and the overall findings, limitations and recommendations reported by the authors of these reviews. In the following two paragraphs we give an overview of the determinant-behavior relationships that have been studied so far, and give an overview of the importance and strength of evidence of potential determinants.

The importance and strength of evidence of potential determinants
With regard to the importance of a determinant and its strength of evidence (Table 6), most determinant-behavior relationships were coded with a zero, indicating that the findings are mixed. The following categories of determinants were found to be significantly related to dietary behavior and/or reported a (non)-significant effect size larger than 0.30 in all identified eligible studies of the included reviews assessing these categories of determinants (++ in Table 3): some aspects of social-cognitive determinants (such as attitude, self-regulation, intention and self-efficacy) and dietary behavior [15,24,27]; habit strength and sugar-sweetened beverage intake [17]; sensory processes and snacking [24]; and sedentary behavior and sugar-sweetened beverage and breakfast consumption [22]. The following categories of determinants were found to be significantly related to dietary behavior and/or reported a (non)-significant effect size larger than 0.30 in more than 75% of the identified reviews assessing these categories of determinants (+ in Table 3): the physical environment and fruit intake [26]; the socialcultural environment and fruit and vegetable intake [16,18,21,24,[27][28][29] and sugar-sweetened beverage consumption [18,20,21,23,24,[28][29][30]; intention, sensory processes, and knowledge for fruit and vegetable intake [24,27]; and sedentary behavior and fruit intake [22], fruit and vegetable intake and snack intake [19,22]. The evidence is mostly limited (limited, suggestive: Ls), predominantly due to the abundance of studies with crosssectional designs so that causal or predictive relations could not be established. Systematic review on the influence of political environments, self-regulation, subjective norm and automaticity were mostly lacking in the included reviews.

Main results
The multitude of studies conducted on determinants of dietary behavior among youth provides mixed and sometimes quite convincing evidence regarding associations between potential determinants and a range of dietary behaviors. However, because of the general use of crosssectional designs in the studies covered in the available reviews, the evidence for true determinants is suggestive at best.
In particular, environmental determinants (mainly the social-cultural environment) and social-cognitive determinants have been studied quite extensively for their association with different dietary behaviors, with somewhat 1) While it was not possible to metaanalyze interaction effects, habit often moderated the relationship between intention and behavior, such that intentions had reduced impact on behavior where habit was strong. This finding must be interpreted cautiously as it may reflect a bias towards publication of studies which find significant interaction, and so an overestimation of the robustness of this effect.
1) Explorations of the role of counterintentional habits on the intention-behavior relationship, such as the capacity for habitual snacking to obstruct intentions to eat a healthful diet, are needed.
2) Healthful behaviors can habituate. The formation of healthful ('good') habits, so as to aid maintenance of behavior change, thus represents a realistic goal for health promotion campaigns.
3) More methodologically rigorous research is required to provide more conceptually coherent and less biased observations of the influence of habit on action.
2) Many studies were cross-sectional, and so modeled habit as a predictor of past behavior. This fails to acknowledge the expected temporal sequence between habit and behavior, and is also conceptually problematic given that, at least in early stages of habit formation, repeated action strengthens habit.

4)
A more comprehensive understanding of nutrition behaviors, and how they might be changed, will be achieved by integrating habitual responses to contextual cues into theoretical accounts of behavior. 3) Reliance on self-reports of behavior. Fruit and vegetable consumption

Use of implementation intentions
Considerable support was found for the notion that implementation intentions can be effective in increasing healthy eating behaviors, with twelve studies showing an overall medium effect size of implementation intentions on increasing fruit and vegetable intake. However, when aiming to diminish unhealthy eating patterns by means of implementation intentions, the evidence is less convincing, with fewer studies reporting positive effects, and an overall effect size that is small. NR 1) Although implementation intention instructions were not included as a moderator in the present metaanalysis due to the limited amount of studies, it seems prudent that future research takes into account the importance of using autonomy supportive instructions.
2) Stricter control conditions as well as better outcome measures are required.
3) Investigate efficacy of implementation intentions in diminishing unhealthy eating behaviors. In doing so, these studies should also compare the efficacy of different types of implementation intentions, as these may have differential effects on unhealthy food consumption. 1) Replication of studies on the same specific environmental factors is necessary, to generate more compelling evidence for associations between environmental factors and dietary intake.
2) The finding that parental behavior is associated with child and adolescent intakes implies that interventions should take the behavior of parents into account, or desensitize adolescents for the (unfavorable) behavior of their parents. Parents should be more strongly encouraged to give the right example, especially 2) Many studies were cross-sectional.

4) Similar environmental determinants
were collapsed conceptually into one category, although potential determinants in the same category were often dissimilar or measured in different ways. This review reported support for three family variables: Parental breakfast eating and living in two parent families were positively associated with adolescent breakfast consumption; and socio-economic deprivation was inversely associated with breakfast consumption.
1) Several studies may not have been powered to detect significant associations between family correlates and breakfast behaviors.
1) Future studies should clearly define breakfast foods (e.g. breakfast cereal, breads, milk, snacks on the run) being measured as this will allow for an understanding of the healthfulness of this behavior and will provide scope for interventions to promote healthy breakfast consumption.
2) Diversity in the definition of breakfast across the literature.
2) Importance of family structure should be considered when designing programs to promote breakfast consumption.
3) Future qualitative studies are needed to further explicate the mechanisms of the complex relationship between SES and adolescent breakfast behaviors.  found for restriction of eating, and an indeterminate result was found for pressuring the child to eat. Food availability was not associated with fruit and vegetable intake and snacking but had an indeterminate result for sweet beverages.
2) Future research should be on interventions to predict whether interventions targeting these correlates will have an impact.
3) Future research should focus on identifying the common correlates of physical activity, sedentary behavior and eating behavior in preschoolaged children so that better tailored interventions could be developed. Fruit and vegetable consumption separately. fruit, fruit juice and vegetable consumption combined.
Children: home availability, family rules (demand/allow) and parental encouragement were positively associated with children's fruit and vegetable intake. Parental modeling and parental intake were positively associated with children's consumption of fruit and fruit juice and vegetable intake. Adolescents: parental intake and parental occupational status were found to be positively associated with adolescents' consumption of fruit. Parental intake was also positively associated with adolescents' vegetable consumption. There is also evidence for a positive association between parental education and adolescents' fruit juice and vegetable intake.
1) Diversity in character (e.g. measures used and correlates studied) 1) More longitudinal studies are needed.
2) More studies are needed to test understudied correlates to generate more convincing evidence for associations between correlates and dietary behaviors. 3) Studies should report the validity and reliability of measures used to assess predictor variables.
2) Difficult to assess overall consistency of associations.
3) Several studies may not have been powered to detect significant associations between family correlates and dietary behaviors. Breakfast consumption, soft drink consumption.
Family and school environment: physical, socio-cultural, economic, political correlates.
Parental descriptive/ injunctive norms and control/supervision were positively related to breakfast. Parental catering on demands, avoidance of negative modeling behavior, permissiveness, and area deprivation were inversely related. School SES was negatively related and teacher injunctive norms was positively related to breakfast. Availability at home, parental soft drink, and permissive parenting style 1) Only studies included that were published in English 1) More longitudinal studies are needed.
2) Interventions could help parents to create a supportive environment for their children to promote healthy behavior.
2) Did not take possible moderators and covariates into account.
3) More research is needed to focus on important school-environmental factors when developing an intervention program.
3) Not all existing studies on this topic were covered.
4) Focused on the consistency of the association and not on the strength of the association. Table 6 Results of the reviews about determinants of dietary behavior among youth (Continued) were positively related to soft drink. Having family dinners, household income, parental employment status, and limits were inversely related. Availability of soft drinks at school and intake at school were positively related with soft drink. Participation in healthy school lunches was inversely related.

5)
Conceptually similar variables were combined into a single category, even if variables were measured in a different way.
Eleven of the 12 included studies reported significant associations between TV and adverse dietary behaviors in young children. Six studies reported significant inverse relationships between TV viewing and fruit and vegetable intake.
1) Reliance on parent-reported methods to assess child TV viewing.
1) Guidelines for TV viewing use in young children should be further delimited. 2) Many studies were cross-sectional.
2) More longitudinal studies are needed.
3) Direct measurement of TV use.
3) Abandon purely distance-based measures of accessibility, and combine multiple environmental assessment techniques.

4)
Researchers should continue to expound upon the conceptual definitions of food access as they develop and refine new combinations of measure for the food environment.
Social status, stress. Higher stress is related to less healthy dietary behaviors. The majority of studies reported that higher social position is related to healthier diet.
1) Only studies included that were published in English.
1) More quantitative dietary assessment tools such as FFQ, repeated 24-hr recalls, and food diaries are needed. 2) Many studies were cross-sectional.
2) More longitudinal studies are needed.
3) Because obesity results from a prolonged period of positive energy imbalance, assessment of dietary behaviors at a single point in time makes inferences related to diet and obesity difficult.
3) Important to acknowledge additional factors that influence energy intake, such as SES and stress levels.
4) Implementing appropriate monitoring and evaluation is essential to identifying successful, holistic strategies that can be used to improve quality of care. In many studies significant associations with general parenting were found. Generally, children raised in authoritative homes were found to eat healthier.

4) Heterogeneity of measures.
1) Reliance on questionnaires and parental self-report measures.
1) Additional research is needed to further study the influence of mediating and moderating factors influencing the general parentingchild weight relationship, preferably employing a longitudinal design with more extended follow-up periods.
2) Differences in conceptualization of parenting constructs across studies.
3) Different categorizations to classify parents into styles across studies.
2) More longitudinal studies are needed using diverse ethnic samples and age groups.  3) Reliance on self-report measures.
2) Beneficial to incorporate mixed qualitative and quantitative designs. 4) Many studies used single informant measures to measure family-level data 3) There is a need for more within-in family measurements that utilize multi-level and multi-measurement approaches. The determinants supported by the greatest amount of evidence are socialeconomic position, preferences, parental intake, and home availability/accessibility. For nutritional knowledge, self-efficacy and shared family meals the evidence for positive associations is rather convincing.
1) Publications may have been missed due to the search strategy.
1) More studies on the influence of the family setting for influencing fruit and vegetable intake among children and adolescents are needed to enable health promoters to make evidence based decisions.
2) Within this review only significant associations are considered.
2) Observational studies analyzing fruit and vegetable intake in a school setting are still lacking.  mixed results. The included reviews suggest that in the past decade, environmental determinants have been studied most extensively. This is an important finding in itself, suggesting a paradigm shift in the field, i.e. from a focus on social-cognitive determinants to environmental factors. This shift towards more consideration of the socialecological approach was also seen in our umbrella review on determinants of dietary behavior in adults [14]. Other potential determinants of dietary behavior, such as automaticity, self-regulation, and subjective norm, have been studied in relatively few studies, but study results are promising. With regard to the outcomes investigated, most reviews explored relations of potential determinants with fruit and/or vegetable intake.

3) Many papers include analyses
In the reviewed papers we found evidence that the social-cultural environment, such as the familial influence (e.g. [21,24,27,28]) is a significant correlate of fruit and vegetable intake and snack consumption in youth in more than 75% of the available studies (see Table 7). Parents, as gatekeepers of the home food supply, can influence children's eating behavior either through the use of specific food parenting practices (i.e. context-specific acts of parenting on child eating including encouraging of food variety and controlling a child's intake of unhealthy products) or through the indirect influence of general parenting [28]. Social-cognitive determinants have been studied often, but the evidence regarding their importance is limited (i.e. suggestive at best). Intention, a proximal indicator of actual behavior, was found to be a significant determinant of fruit and vegetable intake, snack intake, and sugar-sweetened beverage intake [24,27]. Socio-cognitive theories such as the Theory of Planned Behavior [6] are indicated to have limited value in predicting the translation of intention into action. This limitation is addressed in reviews on the constructs of habit [17] and implementation intentions [15]. The review on implementation intentions showed considerable support for the effect of implementation intentions on increasing fruit and vegetable intake (medium effect size) among youth. However, the effect of implementation intentions on the reduction of unhealthy eating patterns was less convincing. Habit strength was one of the factors to be significantly related to sugar-sweetened beverage intake with moderate to strong effect sizes in all identified eligible studies of the review of Gardner et al. [17]. Automatic processes, including habit strength appears to reduce the utility of cognitive factors for the prediction or association with dietary behavior [17]. Additionally, screen time was found to be consistently associated with dietary behavior [19,22,27]. The included reviews provide evidence that the amount of screen time was significantly related to dietary behavior; screen time was positively associated with snack and sugar-sweetened beverage intake [19][20][21][22] and inversely associated with fruit and vegetable intake [19][20][21][22][23][24][25][26][27]. An important mechanism linking screen time to unhealthy dietary behavior is exposure to marketing of unhealthy foods and beverages through screens [35,36]. The food and beverages depicted in these advertisements are predominantly unhealthy foods high in fat, salt and sugar [35,37]. Sedentary activities and unhealthy dietary behavior have repeatedly been found to cluster [38][39][40] and may also share similar environmental cues causing these behaviors to co-occur. Sedentary behavior offers a context for the consumption of energydense food products, disrupting the habituation to food cues.
Systematic reviews on the influence of political environments, self-regulation, subjective norm and habitual behavior were mostly lacking in the included reviews. In addition, some types or categories of potential determinants were not covered in the present umbrella review because we did not come across systematic reviews of such determinants. For instance, although we found two systematic reviews on sensory determinants of dietary behavior [24,27], most of the reviews were excluded as they did not comply to the quality standards of systematic reviews that we used as an inclusion criteria, e.g. [41]. This does not necessarily imply that such factors as taste and preferences are not important, but just that these have not been covered well at present in systematic reviews. Furthermore, it should be noted that lack of evidence for the importance of a possible determinant is not the same as evidence that the determinant is not important; since lack of well-designed studies is often the main reason for lack of evidence. We need to try to distinguish between well-researched determinants and still no evidence for importance, and determinants that have just not been studied (well enough) to make meaningful conclusions.

Limitations and methodological issues
Several limitations should be taken into consideration in reviewing these findings. These include the crosssectional nature of many studies relying on self-report measures; heterogeneity of conceptualization, measurement, samples and analyses used, making it difficult to compare results between studies; inability to conduct a meta-analysis; lack of validity and reliability of dietary intake and correlate measures; and categorization of determinants into more global categories thereby losing important information. Additionally, the systematic reviews included a wide age range, i.e. respondents from birth to 18 years. During childhood many developmental transitions take place that may imply differential importance of distinct behavioral determinants. For instance, parents are highly responsible as gatekeepers of the home food supply for their children's dietary intake behavior. Note: Importance of a determinant: ++, +, 0, −, −-(see Table 3); strength of evidence (see Table 4): Co (Convincing evidence), Pr (Probable evidence), Ls (Limited, suggestive evidence), Lnc (Limited, no conclusion); studies including determinants such as stress and risks and dietary behavior such as milk and meat intake not included in this table.
However, parental influence decreases with advancing age of the child as the child is increasingly exposed to other environments (e.g. school environment, peer influences). In addition to the quality of the research design, the fact that some determinants have not been extensively studied yet, studies of some types of determinants have not been reviewed systematically, and the lack of robust results from this umbrella review may also be explained by the fact that groups or types of determinants are often studied in relative isolation. For instance, studies in which different categories of determinants -e.g. sensory determinants, self-regulation, and political environmental factors -were studied with integrative approaches are largely lacking. Such studies would allow for exploration and testing of mediating and moderating pathways between these determinants in influencing dietary behavior. Already some studies combining environmental and social-cognitive determinants have been reported in recent years and do support such mediating and moderating pathways, e.g. [42][43][44][45].
This is the first umbrella review that provides an overview of reviewed research regarding a broad range of potential determinants of dietary behavior in youth. Umbrella reviews in itself are, however, also prone to bias in various ways. Differences in reviewing methodology and reporting were apparent, as well as differences in for example categorizations of the determinants. By nature, umbrella reviews lead to loss of detail. In addition, some individual studies are included in multiple reviews which may have led to an overrepresentation of single studies in our results. Finally, we excluded reviews that primarily addressed biological determinants or papers with summative outcomes such as caloric intake, and we also did not include reviews that focused on qualitative data.

Conclusions and recommendations
The evidence gathered in our umbrella review suggests that intention and sedentary behavior have the strongest evidence base as determinants of healthy and unhealthy dietary behavior in youth. The influence of distinct determinants may, however, be stronger in interaction with other influences. We would advocate for studies that address combined, mediating and interactive influences on dietary behavior [46]. Such studies are advocated to include behaviors that have been found to cluster with dietary behavior, such as sedentary behavior. Other recommendations include the need for better designed studies, beyond mere cross-sectional research, −i.e. more longitudinal and experimental or intervention research, and research using natural experiments-, larger samples among specific age groups, and more valid and reliable measures (dietary behavior and correlates). Our results underline the importance of embracing theories and factors additional to determinants derived from sociocognitive theories that are often used to inform interventions to promote healthy dietary behaviors. Theories that are promising of further research for determinants of dietary behavior research include habit theory and (social-) ecological models of health behavior.