Conceived and designed the experiments: FNJ. Analyzed the data: FNJ PJK. Wrote the paper: FNJ PJK MB AMdS-S ERL JAP BAS. Interpretation of data: PJK MB AMdS-S ERL JAP. Critical revision for important intellectual content: PJK MB AMdS-S MM ERL JAP BAS. Conception and design of IYM: BAS.
The authors have declared that no competing interests exist.
A number of cross-sectional and prospective studies have now been published demonstrating inverse relationships between diet quality and the common mental disorders in adults. However, there are no existing prospective studies of this association in adolescents, the onset period of most disorders, limiting inferences regarding possible causal relationships.
In this study, 3040 Australian adolescents, aged 11–18 years at baseline, were measured in 2005–6 and 2007–8. Information on diet and mental health was collected by self-report and anthropometric data by trained researchers.
There were cross-sectional, dose response relationships identified between measures of both healthy (positive) and unhealthy (inverse) diets and scores on the emotional subscale of the Pediatric Quality of Life Inventory (PedsQL), where higher scores mean better mental health, before and after adjustments for age, gender, socio-economic status, dieting behaviours, body mass index and physical activity. Higher healthy diet scores at baseline also predicted higher PedsQL scores at follow-up, while higher unhealthy diet scores at baseline predicted lower PedsQL scores at follow-up. Improvements in diet quality were mirrored by improvements in mental health over the follow-up period, while deteriorating diet quality was associated with poorer psychological functioning. Finally, results did not support the reverse causality hypothesis.
This study highlights the importance of diet in adolescence and its potential role in modifying mental health over the life course. Given that the majority of common mental health problems first manifest in adolescence, intervention studies are now required to test the effectiveness of preventing the common mental disorders through dietary modification.
Three quarters of lifetime psychiatric disorders will emerge in adolescence or early adulthood
Data for these analyses were derived from the It's Your Move (IYM) project schools in the Barwon-South Western (BSW) region of Victoria, Australia. This region (population 350,109) covers the south-west coast of Victoria and includes the regional centre Geelong (population 205,929 in 2006). The region has 49 secondary schools with a combined enrolment of approximately 49,000. The IYM project aimed to increase the capacity of schools to promote healthy eating and physical activity, increase the awareness of key messages around active transport and healthy nutrition in homes and early childhood settings, and to evaluate the process, impact and outcomes of the project
Adolescents were sampled in 2005–6 and again in 2007–8. Self-reported information regarding adolescents' key behaviours such as nutrition; mental health and well-being; physical activity; perceptions of the school environment (teachers, canteens, participation in sport); home environment (the role of parents/siblings); neighbourhood environment; and other perception and attitudinal questions was captured with an 84-question survey using Personal Diary Assistants. Surveys were completed during normal class time and took ∼30 minutes to complete. Weight and height were measured by trained researchers.
An ordinal Healthy diet score was constructed from the available dietary data. This scoring method was based on those previously developed and validated in adults
An Unhealthy diet score was constructed using the sum of scores on the following variables: Biscuits, potato chips, other snacks after school; Pies, takeaways or fried foods such as French fries after school; Chocolates, lollies, sweets or ice-creams after school (every day or almost every day = 4; most days = 3; some days = 2; hardly ever or never = 1); plus “In the last 5 school days, how many days did you have non-diet soft drinks?” (0 days = 1 to 5 days = 6); “On the last school day, how many glasses or cans of soft drink did you have? (None = 1 to more than 2 litres = 8); “In the last five school days, how many days did you have fruit drinks or cordials? (0 days = 1 to 5 days = 6); “On the last school day, how many glasses of fruit drinks or cordials did you have? (0 = 1 to 9 glasses = 10); “How often do you usually eat food from a takeaway? (Once a month or less = 1 to Most days = 5); and “In the last five school days, how many days did you buy snack food from shop/takeaway after school? (0 days = 1 to 5 days = 6). The possible range was 9–53, which was subsequently categorised into low (1), medium (2) and high (3) levels of unhealthy food intake based on the distribution of the data.
The Pediatric Quality of Life Inventory (PedsQL) is a pediatric general health profile instrument, developed by Varni and colleagues and specifically designed for use with adolescents and children
Potential confounding factors were identified
Variables included sequentially as potential confounders in the cross-sectional relationship between diet and mental health included age; gender; SEIFA category; dieting behaviours; BMI; and PA. For prospective analyses, examining dietary scores at baseline as predictors of mental health at follow-up, we also adjusted for baseline PedsQL scores. In analyses examining PedsQL scores at baseline as predictors of diet quality at follow-up, diet scores at baseline were also adjusted for. Finally, in analyses regressing change in PedsQL score on change in diet quality, PedsQL scores and diet quality at baseline; and change in PA and change in BMI were also adjusted for.
Descriptive statistics were computed and differences between categories of diet scores tested using linear regression analyses for continuous and chi-square statistics for categorical data. Differences between follow-up (those measured twice) and non follow-up (those measured only at baseline) were tested with t-tests or chi-square tests. Differences between genders on diet scores and change in diet scores were examined using t-tests. Cross-sectional analyses were carried out on the data collected in 2005. Multivariable linear regression analyses were used to examine the cross-sectional associations between diet quality and PedsQL. Two main exposures were used in separate regression analyses: Healthy diet and Unhealthy diet scores. Potential confounders were added to the models sequentially (gender; age; SEIFA category; dieting behaviour; BMI; and PA). Effect modification by age group (less than 15 yr or > = 15 yr), gender, SEIFA scores (high or low) and ‘condition’ (intervention or control group) was also assessed for both cross-sectional and prospective relationships.
Multivariable linear regression analyses also examined the relationship of diet quality in 2005–6 to PedsQL scores in 2007–8. In these longitudinal analyses, variables sequentially adjusted for included PedsQL scores at baseline, in addition to those previously documented. In order to examine the issue of reverse causality, we also examined PedsQL scores at baseline as a predictor of diet quality at time 2. We used the dietary scores at time two as the outcome variables in linear regression analyses, with PedsQL scores at baseline as predictors. Diet quality at baseline was adjusted for, in addition to other listed variables. In final analyses, the variable accounting for change in PedsQL scores was regressed on change in dietary scores. Variables, including those previously tested, as well as change in BMI; change in PA; and both PedsQL scores and diet quality scores at baseline, were also adjusted for.
Final analyses included both Healthy and Unhealthy diet scores as continuous predictors of PedsQL scores in the same model, to test the independence and relative contributions of these constructs. This was done for both cross-sectional and prospective analyses.
Of the initial sample of 3040 students at baseline, 2991 had full data for the PedsQL, 2996 had full data for the dietary intakes and 2921 had full data on relevant covariates. The final sample at baseline consisted of 2915 students with full data available for all three sets of variables (56% males). At follow-up, 2038 had full data for the PedsQL and relevant dietary data and 1958 had full data on relevant covariates. The final follow-up sample, with data available on these three sets of variables, totalled 1949 (54% males). Nearly 60% of the baseline study sample were categorised as “high” SES, based on their SEIFA scores. More than half of students were in the higher category of PA and more than 40% reported dieting behaviour. The majority of students were less than 15 years of age (data not shown). Sample characteristics across categories of diet quality scores at baseline are reported in
|
|
|
|
|
Age | 14.7 (1.4) | 14.6 (1.4) | 14.5 (1.4) | <0.001 |
BMI | 21.7 (3.8) | 21.7 (3.8) | 21.9 (3.9) | 0.77 |
Male | 54 | 58 | 53 | 0.02 |
High Seifa (≥50) | 55 | 60 | 59 | 0.03 |
Dieting behaviour | 43 | 40 | 45 | 0.17 |
High PA (≥median) | 46 | 57 | 63 | <0.001 |
|
|
|
|
|
Age | 14.7 (1.4) | 14.6 (1.4) | 14.6 (1.4) | 0.02 |
BMI | 22.0 (3.8) | 21.9 (3.9) | 21.4 (3.7) | <0.001 |
Male | 44 | 58 | 64 | <0.001 |
High Seifa (≥50) | 61 | 59 | 56 | 0.10 |
Dieting behaviour | 43 | 39 | 44 | 0.03 |
High PA(≥median) | 51 | 54 | 60 | <0.001 |
Results given as Mean (SD) or percentage in each category.
Mean Healthy diet scores were 4.3 (SD±1.7), while mean Unhealthy scores were 20.8 (SD±6.7). Mean Healthy diet scores were similar for males (M = 4.2, SD±6.7) and females (M = 4.3, SD±1.7) (p = 0.27), however mean Unhealthy diet scores were higher for males (M = 21.7, SD±6.7) than females (M = 19.6, SD±6.4) (p<0.001). In cross-sectional analyses, there were positive relationships between Healthy diet scores and PedsQL scores both before and after adjustments (
PedsQL | ||||||
|
|
|
|
|||
|
β-z |
|
|
β-z |
|
|
Least healthy (n = 690) | 0 | 0 | ||||
2 (n = 1716) | 0.36 | 0.27 to 0.45 | <0.001 | 0.31 | 0.22 to 0.39 | <0.001 |
3 (n = 508) | 0.45 | 0.34 to 0.57 | <0.001 | 0.42 | 0.31 to 0.53 | <0.001 |
|
|
|
||||
Least unhealthy (n = 854) | 0 | 0 | ||||
2(n = 1165) | −0.13 | −0.21 to −0.04 | 0.001 | −0.14 | −0.23 to −0.06 | 0.001 |
3 (n = 977) | −0.29 | −0.38 to −0.20 | <0.001 | −0.29 | −0.38 to −0.20 | <0.001 |
*Adjusted for gender, age, dieting behaviours, BMI, SES and PA.
In longitudinal analyses, Healthy diet scores at baseline predicted PedsQL scores at follow-up, both before and after adjustments for gender, age, SEIFA category, dieting behaviours, BMI, PA and baseline PedsQL scores (
PedsQL | |||||||||
|
|
+PedsQL baseline | |||||||
|
β-z |
|
|
β-z |
|
|
β-z |
|
|
Least healthy (n = 690) | 0 | 0 | 0 | ||||||
2 (n = 1716) | 0.26 | 0.15 to 0.37 | <0.001 | 0.22 | 0.12 to 0.33 | <0.001 | 0.11 | 0.01 to 0.21 | 0.03 |
3 (n = 508) | 0.31 | 0.17 to 0.45 | <0.001 | 0.29 | 0.17 to 0.43 | <0.001 | 0.14 | 0.02 to 0.27 | 0.03 |
|
|
|
+ |
||||||
Least unhealthy (n = 854) | 0 | 0 | 0 | ||||||
2 (n = 1165) | −0.00 | −0.11 to 0.10 | 0.97 | −0.01 | −0.11 to 0.10 | 0.82 | 0.05 | −0.04 to 0.15 | 0.27 |
3 (n = 977) | −0.15 | −0.27 to −0.04 | 0.01 | −0.17 | −0.28 to −0.05 | 0.004 | −0.07 | −0.18 to 0.03 | 0.18 |
*Adjusted for gender, age, dieting behaviours, BMI, SES and PA.
When both diet quality constructs were included as continuous variables in the same model, and the model adjusted for all variables other than mental health at baseline, both were significant predictors of PedsQL scores at follow up (Healthy: adjusted β-z = 0.09, 95%CI 0.04 to 0.14, p<0.001; Unhealthy: adjusted β-z = −0.05, 95%CI −0.10 to −0.001, p = 0.045), but these associations were attenuated by final adjustment for PedsQL scores at baseline (Healthy: adjusted β-z = 0.04, 95%CI −0.01 to 0.08, p = 0.12; Unhealthy: adjusted β-z = −0.02, 95%CI −0.07 to 0.03, p = 0.40).
Over the follow-up period, both Healthy and Unhealthy diet scores decreased. However, the decrease in Healthy diet scores was proportionally greater (mean change = −1.2, SD±2.1; 28% reduction in mean Healthy diet score) than the decrease in Unhealthy diet score (mean change = −1.2, SD±6.2; 6% reduction in mean Unhealthy diet score) indicating that diet quality decreased overall. Males demonstrated a greater decrease in Healthy diet scores over the time period (mean change = −1.3, SD±2.1) compared to females (mean change −0.95, SD±2.2) (p<0.001) and a smaller decrease in Unhealthy diet scores over the follow-up period (mean change −0.77, SD±6.6) compared to females (mean change −1.63, SD±5.6) (p = 0.002).
The hypothesis that change in diet quality would be associated with a change in mental health was supported by the data. Improvements in diet quality, as reflected in increases in Healthy diet scores over the two year follow up period, were associated with increased PedsQL scores (adjusted β-z = 0.21, 95%CI 0.14 to 0.28, p<0.001) over the follow-up period, while increases in Unhealthy diet scores were associated with reductions in PedsQL scores (adjusted β-z = −0.13, 95%CI −0.18 to −0.09, p<0.001), both before and after adjustments for previously described covariates, plus change in PA, change in BMI, dietary scores at baseline and baseline PedsQL scores.
Finally, the hypothesis that mental health at baseline would not predict diet quality at follow-up was supported by the data. After adjustments for gender, age, SEIFA category, dieting behaviours, BMI, PA and diet quality at time 1, PedsQL scores at time 1 did not predict diet scores at time 2 (Healthy diet scores time 2: β-z = −0.01, 95%CI −0.06 to 0.03, p = 0.58; Unhealthy diet scores time 2: β-z = 0.03, 95%CI −0.01 to 0.07, p = 0.12). Age group, gender, SEIFA category and/or condition were not identified as effect modifiers of the relationship between diet quality and PedsQL scores in either cross-sectional or prospective analyses.
In this study, diet quality was associated with adolescent mental health both cross-sectionally and prospectively. Moreover, improvements in diet quality were mirrored by improvements in mental health, while reductions in diet quality were associated with declining psychological functioning over the follow up period. Finally, the reverse causality hypothesis, that the reported associations reflect poorer eating habits as a consequence of mental health problems, was not supported by the available data.
These findings from the IYM study are concordant with our previous findings of cross-sectional, dose-response relationships between measures of diet quality and symptomatic depression in Australian adolescents participating in the Healthy Neighbourhoods Study
Another potential explanation is that of unrecognised confounding. In this study we lacked data on familial factors that may promote both poor dietary behaviours and mental health problems in adolescents. Our previous study in Australian adolescents included measures of family conflict and poor family management, as well as dieting behaviours, and did not identify these variables as major confounders in the relationships of interest
It is also the case that the study sample was not necessarily representative of the wider Australian population, being drawn from a population with less cultural diversity
Finally, there were limitations to the data available to construct the dietary scores. We lacked specific information regarding the composition of meals either brought from or consumed at home and it may be that assumptions regarding the quality of these meals were erroneous. However, previous research has shown that diet quality is negatively associated with consuming/purchasing meals outside the home
Data from cohort studies in the UK suggest that the prevalence of emotional and conduct problems in adolescents increased in the period between the mid 1970's and 1999
Paralleling this possible increase in the rates of psychological illness among young people are data indicating a reduction in the quality of adolescents' diets over recent decades. A report based on trends in adolescent food consumption in the US identified a reduction in the consumption of raw fruits, high-nutrient vegetables and dairy foods, which are important sources of fibre and essential nutrients, between 1965 and 1996
There are many pathways by which an insufficiency of healthful foods and/or an excessive intake of unhealthy and processed foods could increase the risk for mental health problems in adolescents. Fruits and vegetables, as well as other components of a healthy diet such as wholegrains, fish, lean red meats and olive oils, are rich in important nutrients such as folate, magnesium, b-group vitamins, selenium, zinc, mono- and polyunsaturated fatty acids, polyphenols and fibre. Many of these nutrients have already been reported as of relevance in depressive illnesses (e.g.
While psychological stress is known to increase the production of pro-inflammatory cytokines, the relationship appears to be bi-directional, with inflammation suggested as a direct contributor to the risk for depressive illness
An important aspect of the shift in habitual diets globally is that of an increase in refined carbohydrate consumption. Hyperglycemia promotes an inflammatory state and high glycemic load (GL) diets are also associated with increased systemic inflammation
This study highlights the importance of diet in adolescence and its potential role in modifying mental health over the life course. Given that adequate nutrition is essential during periods of rapid physical development, and that the majority of mental health problems first manifest in adolescence and early adulthood, intervention studies are now urgently required to test the effectiveness of preventing the common mental disorders through dietary modification. Moreover, the foods available and provided to adolescents need to be receiving much greater attention. Given the findings from this study, particular attention should now be paid to creating environments that promote healthy eating and engaging parents in supporting adolescents to maintain good nutrition during a difficult life stage.
The authors acknowledge the principals, teachers, and students of the intervention schools in the Barwon-South Western region. In particular, the School Project Officers: Sue Blackett, Lee Denny, KerrynFearnsides, Chris Green, Sonia Kinsey, KirstyLicheni, Kate Meadows, Lauren Reading and Lyndal Taylor. Acknowledged also are Colin Bell and others from the ‘Support and Evaluation Team’ at the WHO Collaborating Centre for Obesity Prevention, Deakin University, and Lynne Millar for her valuable work on the IYM project.