The data was sourced from the Netherlands Study of Depression and Anxiety (NESDA) an ongoing longitudinal cohort study designed investigating the course trajectories and consequences of depressive and anxious subjects. The baseline sample consists of 2981 patients [of which 2329 (78%) with a lifetime depressive or anxiety disorder] aged 18–65 years of whom 1979 (66.4%) were female. Patients were recruited in three different Dutch regions from the general population, in general practice and in mental health organisations. General exclusion criteria were an inability to speak Dutch and a primary diagnosis of psychotic, obsessive compulsive, bipolar or severe addiction disorder.
In-depth 4-h interviews in which mental health status, anthropometric measurements, biological measurements and lifestyle factors were assessed at baseline and follow-up which occurred at 2-, 4-, 5- and 9-year intervals. The 9-year assessment included a food frequency questionnaire (FFQ). All participating patients completed written informed consent forms and the research protocol was approved by the Ethical Committee of the participating university. Further details of the NESDA study can be found elsewhere .
The 9-year follow-up assessment was conducted in 2069 persons. For the present study we included those participants who completed the FFQ (n = 1671). Of these, 37 participants were excluded due to improbable energy intake (females: < 500 kcal, > 3500 kcal and males: < 800 kcal, > 4000 kcal ) leaving a total sample of 1634. Of the 9-year follow-up participants, those excluded from the current analyses were more likely to be male, younger, less educated and have a higher severity of depression but not anxiety.
Depressive and anxiety disorder
At each assessment the presence of a DSM-IV depressive [major depressive disorder (MDD), dysthymia] or anxiety disorder (social phobia, agoraphobia, general anxiety disorder and panic) was established using the Composite International Diagnostic Interview (CIDI) version 2.1 . At the 9-year follow-up assessment participants were classified as controls (no lifetime history of depressive or anxiety disorder), current disorder (6-month recency of depressive or anxiety disorders), or remitted disorder (lifetime diagnosis of depressive or anxiety disorder but no current disorder).
Additionally, the severity of symptoms was measured. Depressive symptoms were measured with the 30-item Inventory of Depressive Symptomatology—Self Report (IDS-SR, range 0–84) . The severity of anxiety arousal symptoms was measured using the 21-item Beck Anxiety Inventory (BAI, range 0–63)  and the severity of agoraphobia and social phobia with the 15-item Fear Questionnaire (score range 0–120) .
Dietary intake was assessed with a 238-item, semi-quantitative FFQ which was based on a validated ethnic Dutch FFQ . The FFQ asked about the frequency, amount and type of food eaten in the past month. Using the Dutch Food Composition Table 2014 , daily intakes (g/day) of the 238 food items were calculated. Population medians were imported for missing amounts. Likewise, missing product sort (e.g., full-fat, semi-skimmed or skimmed milk) was replaced with distributions reflecting the population median. The total number of missing items was 1929 (0.6%). The FFQ also included the option to add additional food items consumed within the last week that were not included in the questionnaire. These items were manually re-categorised to comparable food items where possible. Each manual adjustment was made by consensus of two nutritional scientists.
The following 11 food groups (in g/day) were made based on the food groups from the Mediterranean diet score : fruit, vegetables, non-refined grains, legumes, fish, potatoes, olive oil (positively scored), high fat dairy, red and processed meat, poultry (negatively scored). Furthermore, because within the MDS moderate alcohol consumption receives the optimum score and extreme consumptions receive a score of 0, we treated alcohol consumption as a categorical variable. Three categories were non-drinkers (< 36 g ethanol/day), moderate drinkers (≥ 36, < 82 g ethanol/day = reference) and heavy drinkers (≥ 82 g ethanol/day). The overall MDS score was also calculated.
Covariates were selected a priori based on findings from other studies. Gender, age, years of education, partner status (married/living together, single/separated/divorced), smoking status (current, never, former) and physical activity were included as potentially confounding variables. Physical activity during the past week was measured at the 9-year follow-up with the International Physical Activity Questionnaire (IPAQ) and expressed as 1000 MET min/week [31, 32]. Missing values for physical activity (n = 124, 7.5%) were imputed using multiple imputation. Five imputations were made and pooled results of the five separate analysis were used.
Antidepressant used in the previous month were asked during interview and classified according to the Anatomical Therapeutic Chemical (ATC) classification. Use of antidepressants was considered when taken at least 50% of the time.
The analyses were conducted using SPSS 22 (IBM Corp., Armonk, NY, USA). Statistical significance was set at p < 0.05. Socio-demographic characteristics were described using frequencies and means (medians for non-normally distributed variable). Distributions of the 11 food groups, the MDS and total energy intake were also described.
Separate linear regression models were used to estimate the association of energy intake, the MDS and each of the 11 food groups (continuous g/day and alcohol categories), with depression, anxiety arousal and fear severity (continuous standardised IDS, BAI and FEAR, respectively). To mitigate the effect of differential total intakes due to differing energy needs, which varies according to body size, metabolic efficiency, and physical activity, MDS and food groups were adjusted for energy intake using the energy adjustment method . Thus, residuals were calculated for MDS and the 11 food groups by regressing the MDS/food group as dependent variables against total energy intake (kcal/day) as the independent variable. The utilisation of residuals can be conceptualised as the substitution of that particular food group for a similar number of calories from another food source . The residuals from the linear regression analysis were subsequently standardised to enable comparability among food groups, and used in the analyses. As dietary intake and depression are known to be influenced by partner status, the level of education, and other lifestyle factors, we tested two statistical models. The basic model, which included adjustment for age, gender, and years of education, estimates the fundamental relationship between food groups and depression/anxiety accounting for non-modifiable or relatively stable social demographic factors. The fully adjusted model (i.e., age, sex, education, partner status, smoking status, and physical activity) was additionally adjusted for modifiable characteristics. The associations between having current depression or anxiety, or remitted depression or anxiety compared to controls was analysed using multinomial logistic regression analyses. Again, both basic and fully adjusted models were tested. To enable easier interpretation of the magnitude of the relationships between food groups and depression/anxiety, effect sizes were calculated in fully adjusted models using Pearson’s correlations coefficients for linear models and Cohen’s d, defined as the difference in the means between current or remitted diagnosis and controls, divided by the pooled standard deviation of these groups.
To assess the independent effect of any given food groups, a multivariable regression analysis entering all 11 food groups into one fully adjusted model was also performed. Likely, the consumption of certain food groups are correlated with other food groups. Hence, we first examined the correlation (Spearman rho) between food groups and levels of collinearity [variance inflation factor (VIF) and tolerance]. The largest correlation was observed between fruit and vegetables (Spearman rho = 0.34). As the average VIF’s were not substantially above 1, and the maximum VIF was not greater than 10 (max VIF = 1.34) [34, 35] and tolerance levels were not below 0.2 (lowest tolerance was 0.746) we considered multicollinearity not to be a problem. Correction for multiple testing was done for all models using the modified False Discovery rate (Benjamini and Hochberg 1995) method .
To negate the potential effect that antidepressant use may have on food intake, a sensitivity analysis was performed excluding persons taking antidepressants known to affect appetite, namely tricyclic antidepressants (TCA) and mirtazapine .
Finally, the effect modification of the association between the food groups, MDS score and energy intake by sex was examined. However, as no significant interactions (p’s > 0.10) were found the models were not stratified by sex.