Dietary patterns and internalizing symptoms in children and adolescents: A meta-analysis

Context: Studies of child and adolescent internalizing symptoms and dietary pattern have produced mixed results. Objectives: To quantify the association between dietary patterns and internalizing symptoms, including depression, in children and adolescents. Data sources: Embase, PsycINFO, MEDLINE, Web of Science and Cochrane up to March 2021. Study selection: Observational studies and randomized controlled trials with mean age ⩽ 18 years, reporting associations between diet patterns and internalizing symptoms. Data extraction: Mean effect sizes and 95% confidence intervals were determined under a random-effects model. Results: Twenty-six studies were cross-sectional, 12 were prospective, and 1 used a case-control design. The total number of participants enrolled ranged from 73,726 to 116,546. Healthy dietary patterns were negatively associated with internalizing (r = –0.07, p < 0.001, 95% confidence interval [–0.12, 0.06]) and depressive symptoms (r = –0.10, p < 0.001, 95% confidence interval [–0.18, –0.08]). Effect sizes were larger for studies of healthy dietary patterns and internalizing and depressive symptoms using self-report versus parent-report measures, as well as in cross-sectional studies of healthy dietary patterns and depression compared to prospective studies. Unhealthy dietary patterns were positively associated with internalizing (r = 0.09, p < 0.001, 95% confidence interval [0.06, 0.14]) and depressive symptoms (r = 0.10, p < 0.01, 95% CI [0.05, 0.17]). Larger effect sizes were observed for studies of unhealthy dietary patterns and internalizing and depressive symptoms using self-report versus parent-report measures. Limitations: A lack of studies including clinical samples and/or physician diagnosis, and a paucity of studies in which anxiety symptoms were the primary mental health outcome. Conclusion: Greater depression and internalizing symptoms are associated with greater unhealthy dietary patterns and with lower healthy dietary intake among children and adolescents.


Introduction
One in five children and youth worldwide report mental health problems (The Mental Health of Children and Youth in Ontario: A Baseline Scorecard, 2015;World Health Organization, 2003). In particular, internalizing problems, defined as symptoms of depression, anxiety and emotional problems, are frequently reported symptoms among those under the age of 18 years, with increasing proportions of youth endorsing these symptoms over time (Boak et al., 2014). The early identification and treatment of internalizing disorders, such as depression and anxiety, during the child and adolescent years may be critical to prevent ongoing morbidity and mortality from these conditions across the lifespan.
Interest in the role nutrition plays in mental health, known as nutritional psychiatry, has grown in recent years (Adan et al., 2019). Childhood is a time of rapid brain growth, and the impact of specific nutrients, such as dietary omega 3 fatty acids, on cognition has been well-established (Gómez-Pinilla, 2008). There are several hypothesized mechanisms which could explain the link between dietary patterns and internalizing symptoms, including biological explanations (inflammation, gut-brain axis and Brain-Derived Neurotrophic Factor [BDNF]) as well as psychosocial factors. There is also a wide range of ways in which dietary intake (i.e. food frequency questionnaires, diet history questionnaires) and subsequently dietary patterns, including diet quality indices, is characterized in research. In addition, there has been growing interest in considering overall dietary patterns as opposed to single nutrients. This broader approach may help us to better understand associations between dietary patterns and both physical and mental health disease risk compared with the single nutrient approach (Tapsell et al., 2016;World Cancer Research Fund International, 2018). This is because single nutrients and other dietary components are not ingested in isolation, and it is likely that synergistic and potentially antagonistic effects of dietary components influence health and disease risk (Tapsell et al., 2016). Consequently, understanding the effects of overall dietary pattern may have important implications for designing effective interventions to promote health, including mental health.
A recent meta-analysis of 16 studies (n = 45,825) of adults found that dietary interventions reduced depressive symptoms (Firth et al., 2019), suggesting dietary patterns may be an important modifiable risk factor for depression. In this recent meta-analysis of dietary interventions, 'whole-of-diet' interventions were included, as opposed to those focused on individual foods or nutrients, so studies which looked at global measures of diet were selected. While no meta-analysis of literature focused on children and youth has been conducted to date, systematic reviews that have examined these associations in young people have suggested a possible positive association between healthy dietary patterns and improved mental health, as well as unhealthy dietary patterns and poorer mental health (Khalid et al., 2016;O'Neil et al., 2014). In these studies, and consistent with previous research, a 'healthy' diet was defined as one with a greater intake of nutrient dense foods, and an 'unhealthy' diet was noted to have more highly processed foods and sugars, refined carbohydrates and saturated fats. Thus, an 'unhealthy' diet could be considered as less healthy on the spectrum of dietary patterns. Many studies included in these reviews refer to this less healthy dietary pattern as 'Western' in nature (Khalid et al., 2016;O'Neil et al., 2014). That said, empirical findings across the literature have been inconsistent. While clinical best-practice guidelines outline the importance of a healthy eating pattern to improve depression in children and youth (National Institute for Health and Care Excellence, 2018), conflicting data present challenges to clinicians attempting to counsel children and families on specific recommendations for adjustment to diet quality to improve depressive symptoms.
The primary objective of this meta-analysis was to quantify the association between dietary patterns and internalizing problems, defined as symptoms of depression, anxiety and emotional problems, among children and youth under the age of 18 years. Depressive symptoms were highlighted given the abundance of studies looking specifically at these symptoms and diet. Individuals under the age of 18 were also chosen as it is important to examine if the association seen in adulthood between internalizing problems and dietary pattern is present earlier in life, highlighting a potentially modifiable risk factor in the development of potentially lifelong mental health symptoms. To our knowledge, this is the first meta-analysis to examine this association in children and youth. Our secondary objective was to identify potential moderating factors, as these can be particularly informative for clinical interventions, public health policies, and programs related to the promotion of healthy eating and mental health for children and families.

Search strategy
Published studies on internalizing symptoms and dietary patterns were identified by searching the following databases: Embase, PsycINFO, MEDLINE, Web of Science and Cochrane on 20 March 2021. The search was limited to English language articles using keywords related to 'children or youth', 'depression or anxiety or internalizing disorders' and 'diet or nutrition'. These search terms were then combined with the Boolean 'AND'. References of included studies were also searched for additional studies meeting inclusion criteria (see PRISMA Flow Figure 1). No other sources of information were reviewed. The review protocol was registered with the PROSPERO database with reference number CRD42020160405 and is available from https://www.crd.york.ac.uk/prospero/display_record. php?ID=CRD42020160405.

Study inclusion and exclusion criteria
The following inclusion criteria were applied: (1) full-text articles available in English; (2) cross-sectional or longitudinal study design; (3) participant mean age ⩽ 18 years; (4) association between diet and depression and anxiety symptoms as an outcome was provided; and (5) overall dietary pattern, as measured by self-report food frequency questionnaires or another source, was provided. Articles that assessed dietary patterns using only single food items or food groups (i.e. sugar-sweetened beverages only), or that examined individual micronutrients or eating behaviors (i.e. binge eating) exclusively, or employed single-item measures of depression and anxiety symptoms or disorders, were excluded (see PRISMA Flow Figure 1). Australian & New Zealand Journal of Psychiatry, 56 (6) When the same dataset was used in multiple publications, only one study was included. The included study was selected based on the following protocol: if a dataset was used in both cross-sectional and longitudinal studies, the latter was included to maximize the number of prospective studies included in the sample, given the abundance of cross-sectional studies in the literature and the stronger conclusions that can be drawn from longitudinal data. If a single dataset was used in multiple different studies with the same study design, the publication with the largest sample size was included (Higgins et al., 2003).

Data extraction
Titles and abstracts were screened (L.O.), and full-text articles were further assessed for eligibility in an independent manner (L.O. and K.A.S.). Data extraction of full-text articles was performed by two independent coders (L.O. and K.A.S.). A standard coding form was used to collect information on study and sample characteristics. Data were extracted on categorical moderators including study design (cross-sectional or longitudinal), respondent (self-report or parent-report), dietary measure and validity, and type of mental health measure. Data were collected on continuous moderators including study year, sample size, percent boys in the sample, percent of the sample that presented with internalizing symptoms or met criteria for a disorder diagnosis, percent overweight and obese, and percent high socioeconomic status. Adjusted effect sizes were extracted when provided.
In some studies, sample data were stratified by diet quality (i.e. diet quality quartiles) and/or by internalizing symptom status (i.e. positive screen for internalizing symptoms). Results were extracted from the extreme end of the diet quality measure (i.e. most or least healthy and/or unhealthy) and/or from the highest level of internalizing symptom severity.

Quality assessment
Study quality was assessed using the Joanna Briggs Institute Critical Appraisal Checklists for cross-sectional, cohort and case-control studies (Joanna Briggs Institute, 2017a, 2017b, 2017c. Assessment was based on the presence or absence of the following: (1) clearly defined inclusion criteria for the sample, (2) detailed description of the study sample and setting, (3) diet measure had been tested for validity, (4) identification of confounding variables, (5) adjustment for confounding variables, (6) validated mental health measure and (7) appropriate statistical analysis and report of relevant data. Prospective study assessment included acceptable attrition rate (less than 20%), and case-control study assessment included appropriate matching of groups. No follow-up time criterion was applied to prospective studies, as all follow-up times were greater than 6 months. Validity of diet and mental health measures was determined by first checking the included article for indication of measure validity or through a search for publications on measure validity when not indicated in the primary article.

Data synthesis and analysis
Effect sizes were calculated using Comprehensive Meta-Analysis (CMA, 2020) version 3.0 software. Beta values, odds ratios and hazard ratios were entered into CMA. Beta values were entered as correlations into CMA and hazard ratios were entered with p-value estimates for one study. All effect sizes were transformed into pooled correlations (r values) with 95% confidence intervals (CIs) using random-effects meta-analysis, representing the average association between mental health and dietary pattern across studies. Consistent with recent recommendations for psychological research (Funder and Ozer, 2019), pooled effect size magnitudes are interpreted as small, moderate and large based on r values of 0.1, 0.2 and 0.3, respectively.
Heterogeneity of mean effect sizes were assessed using Cochran's Q and I 2 statistics. Cochran's Q is a test of the null hypothesis that the true effect size is shared across studies, and any variance is due to chance (Cochran, 1954). The I 2 statistic is the percentage of total variance across studies due to heterogeneity, as opposed to chance (Higgins et al., 2003). Mixed-effect analyses were used to determine the effect of categorical moderators and meta-regression was used to determine the effect of continuous moderators. Publication bias was assessed using Egger's regression (Egger et al., 1997) and funnel plots. When asymmetry was indicated, Duval and Tweedie's (2000) trim-and-fill analysis was used to estimate adjusted mean effect sizes. Outliers were identified using the boxplot function in R version 4.0.2 (R: The R Project for Statistical Computing, 2020).
A wide spectrum of dietary patterns was defined in the literature. Some variables were continuous, while many studies operationalized the terms 'healthy' and 'unhealthy' based on variables such as the amount of processed or fast foods consumed versus vegetables or complex carbohydrates. During extraction, each measure of diet was identified and summarized individually. To synthesize the information, the measures were then broadly classified as a 'healthy' or 'unhealthy' dietary pattern. Several publications presented separate results for the association between mental health outcomes and measures of healthy and unhealthy diet quality. For this reason, effect sizes were grouped according to whether the diet variable represented healthy or unhealthy diet quality, and meta-analyses were performed separately for these two groups. In publications where fruit and vegetable intake were distinct independent variables, effect sizes were averaged to produce an overall association between a healthy dietary pattern and mental health outcomes. This was done to fit with the study objective of examining the association between internalizing symptoms and overall dietary pattern, rather than individual foods or food groups.
Meta-analyses were first performed with all internalizing mental health measures (depression, anxiety and emotional symptoms). Effect sizes were pooled if studies reported multiple results (i.e. for depression and anxiety), and if results were presented separately for girls and boys. A separate meta-analysis was performed for studies where the primary mental health outcome was depression, as the majority of included studies focused on depression, rather than anxiety or emotional symptoms.

Results
The search of online databases yielded 24,035 total records and 6 additional records were found through examination of reference lists ( Figure 1). After duplicates were removed, the remaining 6880 articles were screened. Following review of title and abstract, 104 full-text articles remained for full-text data extraction by two independent coders. Thirty-nine articles met full inclusion criteria and were included in the meta-analysis.

Study and sample characteristics
Study characteristics. Study characteristics are presented in Table 1. Twenty-six studies were cross-sectional, 12 were prospective, and 1 used a case-control design. The following total number of participants was included in the analyses: 116,546 with respect to healthy dietary patterns and internalizing symptoms; 77,512 in healthy dietary patterns and depressive symptoms; 116,044 in unhealthy dietary patterns and internalizing symptoms; and 73,736 in unhealthy dietary patterns and depressive symptoms. Participant age ranged from 3.9 to 18 years. With respect to age and as outlined in Table 1, the following classification was found in the cross-sectional studies: 1 study of early and middle childhood (birth-year 8, and year 9-year 12); 4 studies of middle childhood (year 9-year 12); and 22 studies included adolescents (year 13-year 18). Of the prospective studies, there were 3 studies of early childhood, 1 study of middle childhood and 7 studies which consisted of adolescents. To collect dietary information, 37 studies used food frequency questionnaires, 2 of which were administered by trained professionals, 1 study used a 4-day diet diary, and 1 study used a 24-hour recall. Seventeen studies assessed both healthy and unhealthy dietary patterns, while 15 assessed only healthy dietary pattern and 7 assessed only unhealthy dietary pattern. The mental health outcome of interest was depression in 24 studies, depression and anxiety in 3 studies, and emotional symptoms in 12 studies. No study examined anxiety as an outcome variable in the absence of depression. Thirty-seven studies used screening questionnaires to assess mental health (28 self-report, 8 parent-report, 1 teacher-report), and 2 studies used physician diagnosis of an internalizing disorder. Dietary intake was determined using self-report in 30 studies, parentreport in 8 studies and teacher-report in 1 study. For studies reporting prevalence rates, the rate of clinically significant depression symptoms ranged from 10.8% to 33.3%, and the rate of potentially clinically significant emotional symptoms in general ranged from 2% to 62.7%.
Study quality. Eighteen studies used dietary intake measures that had undergone validity-testing, 12 of which were assessed in child and adolescent age groups, and 6 in adults aged 19 and older. All studies used validated measures of depressive, anxious or broader internalizing symptoms. The average study quality scores were 5.7 out of 7 for cross-sectional studies, 5.8 out of 8 for prospective studies, and 8 out of 8 for the case-control study. Additional details of the quality assessment can be found in Supplementary  Tables 1 to 3.

Discussion
This meta-analysis found that healthy dietary pattern was significantly associated with fewer internalizing symptoms and particularly depressive symptoms. In contrast, unhealthy dietary pattern was significantly associated with greater symptoms of depression and internalizing symptoms more broadly. All pooled effect sizes are considered small in magnitude and are comparable in magnitude, suggesting both healthy and unhealthy dietary patterns are meaningfully associated with internalizing and depressive symptoms.
The association between depressive symptoms and dietary patterns may be a result of several potential mechanisms, including biological links such as inflammation, gut-brain axis and BDNF, as well as psychosocial explanations. One leading theory that may link diet and mental health symptoms, particularly depressive symptoms, is that of inflammation (Wang et al., 2019), as unhealthy eating may perpetuate a pro-inflammatory state (Perez-Cornago et al., 2014). Among adults, a recent meta-analysis found that a pro-inflammatory diet was associated with an increased risk of depression, especially in women (Wang et al., 2019). Similarly, another recent meta-analysis found that pro-inflammatory cytokines are increased among adolescents with depression and internalizing disorders (Belem da Silva et al., 2017;Colasanto et al., 2020). Although no intervention studies exist in the child and youth populations, lower rates of inflammation after a change in diet were associated with fewer depressive symptoms in one study in adults (Perez-Cornago et al., 2014). Also, diet is also a central factor in how the gut microbiome is configured, with a growing body of literature now demonstrating the association between gut microbiota and mental health disorders (Cryan and Dinan, 2012). A third potential mechanism of association involves a neurochemical, BDNF. BDNF is critical to neuronal functions such as growth and differentiation, and is an important factor in neuroplasticity (Lachance and Ramsey, 2015) with hippocampal levels shown to decrease in response to a highfat, refined sugar diet in animal models (Jacka and Berk, 2007). In terms of psychosocial explanations, it is possible that under periods of stress, individuals are more likely to choose more palatable (high-sugar, high-fat, processed) foods in an effort to self-soothe (Singh, 2014). If and when this becomes a chronic behavior, overconsumption of calorie-dense foods may contribute to depression through biological and psychosocial pathways (Singh, 2014). Finally, another psychosocial pathway to consider would be the impact of food insecurity in a family, which has been found to be predictive of higher levels of children's mental health conditions (Melchior et al., 2012). Thus, there are several psychosocial variables with respect to the relationship between eating and mental health that should be considered, including poverty and food insecurity, availability of parents and families eating together, and the family culture around the enjoyment of food.
In the current meta-analysis, participant sex was a significant moderator in studies of healthy dietary patterns and internalizing symptoms, with studies with higher percentages of male participants having stronger effect sizes.  In one of the few prospective studies available, Trapp et al. (2016) observed that females consuming a Western dietary pattern at 14 years were more likely to have externalizing behaviors within clinical threshold at 17 years (Trapp et al., 2016). Although these findings were not aligned with those of the current meta-analysis, it suggests that there may be a need to consider possible sex differences in future research. In addition, participant body weight (overweight or obese) was also a significant moderator, with higher percentages of overweight and obese participants associated with stronger effect sizes in studies of unhealthy dietary patterns and symptoms of internalizing problems. Studies of internalizing symptoms and weight are mixed. Prospective studies have found internalizing symptoms to be associated with later weight gain (Camfferman et al., 2016), and depression in youth associated with higher body mass index (BMI) in adulthood (Korczak et al., 2014). However, a cross-sectional study found no association between emotional problems and weight status among preschool children (Mackenbach et al., 2012). Taken together with the findings in the current study, research suggests the association between body weight and mental health is likely bidirectional, with greater evidence for mental health status influencing later weight gain.
In the current meta-analysis, associations were stronger for studies using child-and adolescent-report, rather than parent-or teacher-report, of dietary intake, for studies of internalizing symptoms and healthy and unhealthy dietary patterns. Informant discrepancy between parents and children has been documented in studies of dietary habits, with one study finding that compared to parents, children tended to report greater allowance of soft drinks and fruit juice, less encouragement to eat breakfast and more availability of soft drinks at home (Rebholz et al., 2014). Children and adolescents may be more aware of their own intake and thus more accurate in reporting on what they eat. Parents may not know what types of foods children consume outside of supervision (e.g. at school). It is also possible that children and adolescents are more willing than their parents to report on certain dietary behaviors, such as the amount and frequency of junk food consumption, as parent responses may be influenced by social desirability (Karver, 2006).
Associations were also stronger for cross-sectional versus longitudinal studies examining depression and/or internalizing symptoms and healthy dietary pattern, with non-significant associations between healthy dietary pattern and future depression and/or internalizing symptoms. In the current study, the pooled association between healthy dietary patterns and depressive symptoms was larger than that of healthy dietary patterns and depressive and/or internalizing symptoms, suggesting the relationship between diet and mental health to be stronger for depression than other forms of internalizing symptoms. Although there is evidence in adult populations that improvements to dietary pattern can improve symptoms of depression (Firth et al., 2019), the reverse may also be true, such that children and adolescents with depression are less likely to engage in healthy eating behaviors. Concurrent associations may be more robust given the tendency for changes in appetite or attitudes toward foods to occur in conjunction with mood difficulties. Indeed, emotional eating-eating in an effort to cope with negative emotions-has been associated with depressive symptoms in adolescent populations (Hou et al., 2013). Furthermore, associations between depression and eating disorder symptoms (e.g. eating large amounts of food in the absence of physical hunger) have been shown to increase in magnitude during adolescence, suggesting that depression and disordered eating behaviors may develop concurrently during youth (Abbasalizad Farhangi et al., 2018). The directionality of associations therefore warrants greater exploration in future research given its implications for targets of intervention.
Dietary pattern may be an important modifiable risk factor for children and youth with respect to symptoms of anxiety and depression. Clinical best-practice guidelines outline the importance of a healthy eating pattern to improve depression in children and youth (National Institute for Health and Care Excellence, 2018). This line of research could help inform clinicians working with youth with depression and other internalizing problems to consider a referral to a dietician. In addition, these findings would also support the need for dieticians to be trained in working with children and adolescents with depression and other internalizing problems. Indeed, other primary care and children mental health clinicians may similarly benefit from dietary education and/or collaboration with dieticians, so that mood and diet concerns can be addressed concurrently.

Limitations
Several limitations should be considered. First, only one study included in the meta-analysis used physician diagnosis of an internalizing disorder to measure mental health, the remaining used self-report or parent-report measures. Moreover, the majority of studies were of community populations, limiting the generalizability of our findings to clinical samples. Future research should focus on associations between dietary patterns and mood in clinical samples to add to our understanding of this association in the most severely affected and vulnerable populations, as well as circumvent issues of reporting and/or recall bias that may be increased in self-report measures of mental health. Similarly, the majority of studies used food frequency questionnaires. Future studies may achieve greater accuracy of dietary intake through use of more comprehensive measures, such as online recalls. As well, typically the methods used to characterize dietary patterns are not able to account for potential synergies in intake but rather examine the way in which intake of specific foods cluster together or align with an a priori notion of healthy eating, based on an index. Second, only two included studies examined anxiety and dietary patterns, establishing a need for further research on diet and anxiety independent of depression in adolescent populations. Also, it should be considered that the ability to eat in a way which could be characterized as 'healthy' may also indicate other measures of health, impacting emotional functioning, such as home environment. Although several moderators were explored, studies were inconsistent in reporting variables including socioeconomic status, bodymass index and exposure to stress as measured through markers of inflammation or physiological mechanisms, reducing the number of studies that could be included in moderator analyses examining these potentially important factors overall. The inclusion of these variables in future research would be important in further defining the relationship between dietary pattern and internalizing symptoms. Finally, as noted in Tables 4 and 5, age was not found to be a continuous moderator. There were limited studies in the early to middle groups. This finding illustrates the need for research among pre-adolescent-aged children.

Conclusion
This meta-analysis finds that depression and internalizing symptoms are positively associated with unhealthy dietary pattern, and negatively associated with healthy dietary pattern among children and adolescents. The study also highlighted several important moderators. The dietary informant, participant sex and weight status were found to be important variables and may be considered for future research in this area, as well as potential targets for intervention. Future research is also required to establish the direction and mechanism of the association between diet and depression among children and adolescents and to identify the role of dietary improvements in the prevention and treatment of these disorders, in order to impact both mental and physical health outcomes.
Dr Orlando conceptualized and designed the study, assisted in data collection and extraction, and assisted in drafting the initial manuscript and revised the initial manuscript; Ms. Savel participated in data collection and extraction, conceptualized and performed the initial analyses, drafted the initial manuscript and revised the manuscript; Dr Madigan provided input into the method and analyses, contributed to and reviewed the manuscript; Ms. Colasanto assisted in feedback around method and analyses, contributed to and provided important feedback regarding the manuscript; Dr Korczak participated in conceptualization and design of the study, provided critical feedback around the initial analyses and methods, as well as contributed to and provided important feedback regarding the manuscript and revisions. All authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.