To investigate the association between dietary patterns and prevalence and incidence 3 years later of depressive symptoms using data from the mid-aged cohort in the Australian Longitudinal Study on Women’s Health.
Participants (aged 50–55 years) completed a food frequency questionnaire in 2001. Depressive symptoms were measured in 2001 and 2004 using the validated 10-item Centre for Epidemiologic Studies Depression scale. Multiple logistic regression was used for cross-sectional analysis (8369 women) and longitudinal analysis (7588) to assess the associations between dietary patterns and prevalence of depressive symptoms, and then for longitudinal analysis (6060) on their associations with the incidence of depressive symptoms in 2004, while adjusting for sociodemographic and lifestyle factors.
Six dietary patterns were identified from factor analysis: cooked vegetables, fruit, Mediterranean style, meat and processed meat, dairy, and high fat and sugar. A higher consumption of the Mediterranean-style diet had a cross-sectional association with lower prevalence of depressive symptoms in 2001, adjusted odds ratio 0.82 (95% confidence interval 0.77–0.88); and longitudinally with lower incidence of depressive symptoms in 2004, adjusted odds ratio 0.83 (0.75–0.91). None of the associations found for other dietary patterns remained statistically significant after adjustment for confounders. A dose–response relationship was found cross-sectionally when women were grouped according to quintiles of Mediterranean-style diet (P-value for trend <0.001).
Consumption of a ‘Mediterranean-style’ dietary pattern by mid-aged women may have a protective influence against the onset of depressive symptoms. These findings suggest that dietary patterns have a potential role in the prevention and management of depressive symptoms.
Depression is expected to become the world’s second leading disease burden, after cardiovascular disease, by 2020.1 People with depression report poorer quality of life, affecting their social functioning and performance at work.2 Commonly used therapies to treat depression include the prescription of pharmaceutical antidepressants and psychotherapy,3 but nutrition has also been suggested as having an antidepressant role. Several observational studies have investigated the intake of single nutrients, such as folate, fatty acids, vitamins B6 and B12 in relation to depression.4 However, because diets comprise many nutrients that are likely to interact with each other, it is also appropriate to consider the role of dietary patterns.
Most observational studies that have investigated the association between dietary patterns and depression have been cross-sectional, and therefore have provided limited information on the causal pathways.5, 6, 7, 8 For instance, associations might reflect reverse causation, whereby having poor diet maybe one of the consequences of depression. A number of prospective studies have examined dietary patterns and subsequent depression and have identified an inverse dose relationship for ‘whole food intake’9 and Mediterranean dietary pattern score.10 Few studies, however, have addressed the relationship of dietary patterns and depression among women (for instance, less than a quarter and a third of the sample, respectively, were women in the two previous studies), even though the burden of depression among women is estimated to be twice that of men.11 Where studies have focussed on women, they have examined dietary patterns and specifically postpartum depression in mothers,12, 13 or the association between specific nutrients and clinical depression.14
The aim of this prospective study of women is to investigate the association between dietary patterns, obtained by factor analysis of data from a food frequency questionnaire (FFQ), and the incidence of self-reported depressive symptoms after 3 years. It uses prospective data from mid-aged cohort of women born in 1946–1951 and participating in the Australian Longitudinal Study on Women’s Health. It was hypothesised that a healthy diet, rich in fruit, vegetables and fish, would be associated with a lower incidence of depressive symptoms.
Materials and methods
The Australian Longitudinal Study on Women’s Health is a prospective cohort study investigating multiple factors affecting the health and well-being of more than 40 000 women, who were aged 18–23, 45–50 and 70–75 years at baseline in 1996.15 Each of the three cohorts has been followed up with self-administered questionnaires at 2–4-year intervals. The surveys include questions about social support, quality of life, demographics and health behaviours, life events, diagnoses, symptoms and health-service use.16 Participants were randomly selected from the Medicare health insurance database, which includes all Australian citizens and permanent residents, with oversampling of women living in remote and rural areas. Further details about the study are described elsewhere15, 16 and can be found from the study website (http://www.alswh.org.au).
This study focuses on the mid-aged cohort and data collected at survey 3 in 2001 and survey 4 in 2004. The baseline survey in 1996 had an estimated response rate of 53–56%, and the respondents (n=13 716) were shown to be broadly representative of the Australian population.16 The response rates for surveys 3 and 4, respectively, were 84.3% (n=11 226) and 84.0% (n=10 905) of the women who had completed survey 1 and had not died or dropped out because of ill health. The study methods are approved by the Human Research Ethics Committees of the Universities of Newcastle and Queensland.
Depressive symptoms experienced during the week before the survey were identified using the 10-item Centre for Epidemiologic Studies Depression Scale at survey 3 and 4. This is a self-reported scale that is well validated and has good test retest reliability and predictive validity compared with the original 20-item version.17, 18, 19 Responses were reported using a four-point Likert scale ranging from ‘rarely’ (scored 0) to ‘most of the time’ (scored 3). Possible scores range from 0 to 30, with higher scores indicating more depressive symptoms. A score of 10 was used to define presence of depressive symptoms.17, 18 For the sensitivity analysis, the definition of depressive symptoms was expanded to include women using antidepressant medication.
Dietary intake was assessed at survey 3, using a FFQ developed for use in Australian adults.20, 21 This questionnaire asks participants to report their usual consumption of 74 foods and six alcoholic beverages over the last 12 months using a 10-point frequency scale ranging from ‘never’ to ‘three or more times per day’. In addition, the FFQ includes questions about the type of milk, bread, fat spreads and cheese used and the amount of milk, bread, sugar, eggs and cheese consumed. Responses to the total of 101 items of foods and beverages were converted to daily frequencies. This questionnaire has been previously validated against 7-day weighted food records in a sample of 63 women.20 Total energy intakes were calculated from NUTTAB 1995, the Australian food composition table.22
Other explanatory variables
Numerous potentially confounding variables were considered based on previous studies that have investigated dietary patterns and depression. The sociodemographic variables were: age, area of residence categorised according to an index of distance to the nearest urban centre (urban, rural and remote), ability to manage on available income (impossible/always difficult, sometimes difficult, not too bad and easy), occupation (categorised using the Australian Standard Classification of Occupations),23 education (no formal qualification/school certificate, higher school or trade certificate/diploma and university degree) and marital status (married/de facto, separated/divorced, widowed and never married). Lifestyle variables were: smoking (never smoked, ex-smoker, light smoker <10 cigarettes per day, moderate smoker 10–20 cigarettes per day, heavy smoker >20 cigarettes per day) and physical activity (nil/sedentary, low, moderate and high).24 Health-status variables were: body mass index using the World Health Organisation classification (underweight <18.5 kg/m2, normal weight 18.5–24.99 kg/m2, overweight 25–29.99 kg/m2 and obese 30 kg/m2), change in body mass index in kg/m2, total energy intake in kJ, history of non-insulin-dependent diabetes mellitus (yes/no), hypertension (yes/no), heart disease (yes/no), stroke (yes/no) and mean stress score. This last scale measures stress using responses to items in specific life domains: own health, health of other family members, work/employment, living arrangements, study, money, relationship with parents, partner/spouse, children and anything else.25 All these variables were reported at survey 3, except educational attainment was reported at survey 1 and change in body mass index, which was assessed between survey 3 and 4.
The cross-sectional analysis included all participants with valid data on diet and depressive symptoms at survey 3 (n=8369). From this group, participants were only excluded from the first longitudinal analysis (n=7588) if they had missing depressive symptoms data at survey 4. For the last analysis of incident cases of depressive symptoms at survey 4, women were excluded if depressive symptoms were present at survey 3 (n=1528), leaving a sample of 6060 women (Figure 1).
The statistical analysis was carried out in five steps. First, dietary patterns for all women with dietary data (n=10 629) were derived using factor analysis on the 101 food items with varimax rotation.26 Six factors were identified based on having an eigenvalue >2 and the shape of the scree plot.26 Foods with a factor loading 0.30 were considered important contributors to each dietary pattern and were used for labelling the factors. Factor scores for each dietary pattern were calculated for each participant using all the food items. Second, demographic, lifestyle and health-status characteristics between four groups of women defined by presence or absence of depressive symptoms at survey 3 and 4 were compared by using analysis of variance for continuous variables and χ2-tests for categorical variables.
Third, cross-sectional analysis was carried using multiple logistic regression to estimate the odds ratios and 95% confidence intervals for depressive symptoms for all women at survey 3 with respect to each of dietary patterns identified from step one, while adjusting for other confounders. Fourth, using the same approach as Kesse-Guyot et al.,27 longitudinal analysis was undertaken to investigate depressive symptoms at survey 4, and the analysis was repeated with additional adjustment for those women who reported depressive symptoms at survey 3. In the final step, using only data from women without depressive symptoms at survey 3, multiple logistic regression was again used to estimate the association between depressive symptoms at survey 4 and each of the six dietary patterns, while adjusting for other confounders. Where an association was detected, a linear trend test using quintiles of factor scores was undertaken. All analyses were performed using STATA version 11 (StataCorp LP, College Station, TX, USA), and because of the large sample size and multiple comparisons, the significance level was set at 0.1% (P<0.001).
Six dietary patterns were identified at survey 3 from factor analysis (Table 1) and were labelled: ‘cooked vegetables’ (includes cauliflower, cabbage, Brussels sprouts, broccoli and green beans), ‘fruit’ (strawberries, pineapple, melon, apricots and mango), ‘Mediterranean style’ (garlic, peppers, mushrooms, salad greens, pasta and red wine), ‘meat and processed meat’ (pork, bacon, sausages and lamb), ‘dairy’ (cream cheese, low-fat cheese, yoghurt and skim milk) and ‘high fat and sugar’ (sweet biscuits, cakes, jam, meat pies and chocolate).
Main characteristics of participants by depressive symptoms status at survey 3 and 4 are shown in Table 2. Of the 7588 participants, 70.3% reported no depressive symptoms at survey 3 and 4, 9.5% had symptoms at survey 3 but not at survey 4, 8.7% had symptoms at survey 4 but not at survey 3 (incidence) and 11.5% reported symptoms at both surveys. Compared with this last group of women, those without depressive symptoms at both surveys were more likely to manage on their available income easily, be in managerial/professional occupations, have higher education levels, be married, never have smoked, be physically active, have normal weight and lower stress scores.
The ‘Mediterranean style’, ‘meat and processed meat’ and ‘dairy’ dietary patterns were found to have cross-sectional associations with depressive symptoms (Table 3). However, only the strong inverse cross-sectional relationship of the ‘Mediterranean-style’ diet was unattenuated after adjusting for confounders. This inverse relationship for the ‘Mediterranean-style’ pattern was also apparent in the longitudinal analysis and remained after adjusting for confounders and depressive symptoms at survey 3: adjusted odds ratio per unit of factor score 0.85 (95% confidence interval 0.79–0.91). In both the cross-sectional and the longitudinal analyses, the positive association with ‘meat and processed meat’ and the inverse association with ‘dairy’ dietary patterns were attenuated after adjusting for smoking status and physical activity levels.
Using data only from women without depressive symptoms at survey 3 (Table 4), a similar inverse association was identified between the ‘Mediterranean-style’ diet at survey 3 and the incidence of depressive symptoms at survey 4: unadjusted odds ratio per unit of factor score 0.82 (95% confidence interval 0.75–0.89). The relationship was not attenuated after adjustment for confounders: odds ratio 0.83 (0.75–0.91). For the ‘dairy’ dietary pattern, an inverse association was found, odds ratio 0.86 (95% confidence interval 0.79–0.94). After adjustment for confounders, in particular smoking status and physical activity levels, however, the association was no longer statistically significant. When women were grouped according to quintiles of adherence to the ‘Mediterranean-style’ diet based on intake scores, a dose–response relationship was found (P-value for linear trend <0.001). Those who had had highest intake were 37% less likely to report depressive symptoms, (adjusted odds ratio 0.63, 95% confidence interval 0.47–0.85) compared with the quintile of women with the least adherence to the ‘Mediterranean-style’ diet (Figure 2).
Several sensitivity analyses were conducted according to these scenarios: dietary patterns were derived from the analysis sample (n=6060); excluding women who were also depressed at survey 2 in 1998 (n=738), the survey before the study period; presence of depressive symptoms, defined as those who were on antidepressants or had a score of 10 on the Centre for Epidemiologic Studies Depression Scale; and the use of multiple imputation to deal with missing covariates. Women with diabetes, hypertension, heart disease and stroke at baseline were then excluded. Finally, those with low (<2093.5 kJ) or high (14654.5 kJ) energy intake were also excluded (Table 5). In all scenarios, a similar association was found between ‘Mediterranean-style’ diet and incidence of depressive symptoms.
This prospective study investigates the association between dietary patterns and depressive symptoms 3 years later among a cohort of mid-aged women. We identified six dietary patterns, including a ‘Mediterranean-style’ dietary pattern characterised by the intake of garlic, peppers, mushrooms, salad greens, pasta and red wine. After adjusting for sociodemographic and lifestyle factors, the ‘Mediterranean-style’ dietary pattern was found to have both a cross-sectional and prospective relationship, with the latter showing a 17% reduction in the odds of reporting incidence of depressive symptoms 3 years later. A strong dose response was also evident, whereby women in the highest intake quintile of the ‘Mediterranean-style’ pattern were less likely to report subsequent depressive symptoms by more than a third, odds ratio 0.63 (95% confidence interval 0.47–0.85) compared with women in the lowest quintile of intake.
Comparison with other studies
A direct comparison with findings from other studies is difficult, because of differences in study designs, the methods used to derive dietary patterns and the measures of depression. A recent prospective study of UK office workers found some evidence that a higher consumption of the ‘whole food’ diet, characterised by high intakes of vegetables, fruits and fish, was associated with fewer depressive symptoms 5 years later.9 In another prospective study of Spanish graduates, increasing adherence to a Mediterranean dietary pattern (scored on high intake of fruits, vegetables and fish, moderate alcohol consumption and low intake of meat and dairy products) was prospectively associated with lower risk of clinically diagnosed depression.28 A recent French study found a cross-sectional association of n-3 polyunsaturated fatty acid intake and the prevalence of depressive symptoms, but no longitudinal association was detected with the incidence of symptoms 13 years later;27 a considerably longer period between dietary assessment and health outcomes than the 3-year follow-up period in our study. Similar results from cross-sectional studies have been found, for example, for a ‘healthy Japanese’ diet consisting of vegetables, fruit and soybeans.5 Although broadly consistent with our findings, these studies do not focus on dietary patterns and depression among a population-based sample of women.
In a cross-sectional study of women from the Geelong Osteoporosis Study, a ‘traditional’ diet, comprising vegetables, fruit, meat, fish and whole-grain foods, was associated with a lower risk of depression.8 Samieri et al.7 used cluster analysis to derive dietary patterns and found that elderly women in Bordeaux, France, reported fewer depressive symptoms when they consumed a ‘healthy’ (raw and cooked vegetables, fruit and fish) or ‘charcuterie and starchy foods’ (meat, preserved meats, pasta, rice, potatoes and bread) diet. In prospective studies on women’s health, a recent Japanese study was unable to identify clear associations between dietary patterns obtained from factor analysis, including the ‘healthy’ and ‘Japanese’ dietary patterns, and the subsequent risk of postpartum depression.12 Another study from Crete, however, found that women with the highest quartile of consumption of a ‘health conscious’ diet, which was essentially a Mediterranean-style dietary pattern, were at significantly lower risk of postpartum depression.13 Although findings from our study are consistent with these results, they apply to broader measure of depression than just the incidence of postpartum depression.
A high intake of processed foods was positively related to depressive symptoms in office workers,9 and in adolescents, a Western dietary pattern, including take-away foods, confectionery, red meat and cakes, was associated with poorer mental health.6 Findings from the prospective study of Spanish graduates study have also shown that the risk of depression increased with the consumption of fast-food items (hamburgers, sausages and pizza).29 An association between intake of fast-food and depressive symptoms was found in a cross-sectional study of US women.30 Depressive symptoms were also positively associated with ‘biscuits and snacking’ among elderly French women,7 and among Australian women consuming a ‘Western’ dietary pattern (meat pies, processed meat, pizza, chips and hamburgers),8 but these results were no longer statistically significant after adjustment for other variables. Although we did not identify an association between an unhealthy diet and depressive symptoms, this maybe because of the larger number of dietary patterns we examined compared with the other studies.8, 9 In our study, separate dietary patterns for ‘meat and processed meats’ and ‘high fat and sugar’ were identified, while in other studies these form a single dietary pattern,8, 9 composed of more food items and consequently greater statistical power to detect an association.
Strengths and weaknesses
The strengths of this study lie in the use of a large community-based longitudinal study of women, the use of validated measures for self-reported dietary intake and depressive symptoms, and the ability to adjust for a wide range of sociodemographic and lifestyle factors. Although adjustment for potential confounders had little effect on the association between ‘Mediterranean-style’ dietary pattern and the prevalence and incidence of depressive symptoms, residual confounding could still be present as there maybe unmeasured (such as the family history of depression and personality traits) or imprecisely measured confounders (such as physical activity). Unlike previous research on diet and depression, the findings from this study, including the impact of sociodemographic and lifestyle factors, are specific to women, which is important as women in Australia and women globally have higher rates of depression than men.31
Our use of factor analysis as a data-driven approach to identify dietary patterns from FFQs is also a strength as every participant obtains a score for each dietary pattern, in contrast to typical cluster analysis in which individuals are characterised by mutually exclusive dietary patterns.32 Factor analysis, however, still involves several arbitrary decisions, the effects of which we reduced, for instance by not collapsing the food items into a smaller number of food groups and using the scree plot.32 Also, some items of interest, such as intake of olive oil, were not specified in the FFQ.
We undertook a number of sensitivity analyses to address potential weaknesses with regard to reproducibility of dietary patterns and robustness of the model. The dietary patterns obtained in this study might not be reproducible across other studies as a consequence of differences in FFQs. Nevertheless, we repeated the factor analysis and the logistic regression on the complete cases (n=6060), and findings remained effectively unaltered.
The analysis of this study has been structured to reduce the possibility of reverse causality—whereby depression affects diet—as an explanation of our findings. We found an inverse dose response for Mediterranean-style dietary pattern scores as survey 3 and the reporting of depressive symptoms (Table 2): with women who reported depressive symptoms at survey 4 and women who reported depressive symptoms at surveys 3 and 4, both having lower scores than other women. We additionally excluded from the analysis those women who reported depressive symptoms at survey 2. We also repeated the analysis excluding women with diabetes, hypertension, heart disease and stroke at baseline, and again excluding those with low or high energy intake. In all cases, this had no substantive effect on the association between ‘Mediterranean-style’ dietary pattern and depressive symptoms.
Meaning of the study
A number of plausible biological mechanisms have been suggested for the association between diet on depression, including some that are consistent with a preventive role for a Mediterranean-style diet. Diet might affect brain functions that are involved in the aetiology of depression, including synthesis and regulation of neurotransmitters,33 synaptic plasticity,33, 34, 35 membrane fluidity and neuroinflammation.34, 36 Depression has been associated with low levels of the neurotransmitter serotonin.35, 37 The only precursor for serotonin is tryptophan,35, 37 which is an essential amino acid. Dietary sources include fish, legumes, whole grains and nuts.
Another plausible mechanism relates to the ratio of omega-6 to omega-3 fatty acids in Western diets. Omega-6 fatty acids are associated with an increase of proinflammatory eicosanoids, a decrease of brain-derived neurotrophic factor and a decrease in membrane fluidity.3, 36 It has been shown that in people with depression, inflammation is increased34, 38 and brain-derived neurotrophic factor is reduced.35 The omega-3 fatty acid docosahexaenoic acid, which is most abundant in the brain, can inhibit the release of proinflammatory cytokines34, 36 and increase the levels of brain-derived neurotrophic factor, which in turn enhances neurotransmission and synaptic plasticity.33, 34, 35
Furthermore, increased homocysteine levels have been associated with depression, and folate deficiency could be one of the causes.39 Folate is present in dark-green leafy vegetables, legumes, citrus fruits and many types of bread and other cereals (which are fortified in Australia). As a consequence of low folate levels, the synthesis of homocysteine to methionine and S-adenosyl-methionine is impaired. S-adenosyl-methionine is a methyl donor and is involved in the synthesis and metabolism of neurotransmitters.39
In summary, the results indicate that consumption of the ‘Mediterranean-style’ dietary pattern by mid-aged women may have a protective influence against the onset of depressive symptoms 3 years later. It suggests that health professionals and public policymakers should consider diet as having a potential role in the prevention and management of depressive symptoms.
We thank Professor Graham Giles of the Cancer Epidemiology Centre of The Cancer Council Victoria for permission to use the Dietary Questionnaire for Epidemiological Studies (Version 2), Melbourne: the Cancer Council Victoria, 1996. We also thank all the participants for their valuable contribution to this project. The Australian Government Department of Health and Ageing provided funding for the Australian Longitudinal Study on Women’s Health. GDM is supported by the Australian National Health and Medical Research Council.
About this article
European Journal of Nutrition (2018)