Between August 1991 and March 1992, a random sample of 96,000 women at age 29-49 and residing in the Uppsala healthcare region of Sweden was drawn from the Total Population Register by Statistics Sweden. These women were invited to participate in the Women’s Lifestyle and Health (WLH) study, through answering a comprehensive questionnaire including food frequency questionnaire (FFQ) at the cohort entry [20]. Among the women invited, 49,261 returned the questionnaire and were enrolled in the study. All eligible participants were followed from cohort entry as defined by return of the questionnaire in 1991/92 until the first diagnosis of depression, date of emigration, death, or December 31, 2012, whichever came first, through cross linkages to the Swedish Patient Register (for ascertainment of depression) and Total Population Register (for ascertainment of emigration and death), using the individually unique personal identity numbers [21].
Adherence to MDP
At cohort entry, the study participants were asked to recall their dietary habits during the six months before enrolment, through answering an FFQ, which assessed the frequency and quantity of consumption of approximately 80 food items and beverages [22]. The consumption (grams/day) of each item and total energy intake (kilo-Joule (kJ)/day) were calculated using the Swedish National Food Administration database [23]. To measure adherence to MDP, we used the scale proposed by Trichopoulou et al. [9, 24] which has been used extensively in different studies including the WLH study [25].
Altogether, nine food groups were constructed as index dietary components, namely vegetables, fruits and nuts, cereals, legumes, dairy products, fish and seafood, meat, alcohol, and monounsaturated-to-saturated fat (M/S) ratio. For dietary components that are presumed to be beneficial (i.e., vegetables, fruits and nuts, cereals, legumes, fish and seafood, and a high M/S ratio), we scored a woman that consumed below the median level of the entire cohort as “0” and a woman that consumed at or above the cohort median as 1. For dietary components that are presumed to be less beneficial (i.e., dairy and meat products), a consumption level below the cohort median was given a score of 1 whereas a consumption level at or above the cohort median was given a score of 0. A moderate level of alcohol consumption (5-25 g/day) was scored 1, or 0 otherwise. Scores on all nine components were then summed up as a proxy for adherence to MDP, with the value 0 as the minimal and 9 as the maximal adherence [24].
Diagnosis of depression
The outcome of the study was the first clinical diagnosis of depression during follow-up. The Swedish Patient Register includes nationwide complete information on inpatient psychiatric care since 1973 and outpatient specialist care since 2001, updated on a daily basis [26]. A clinical diagnosis of depression was identified using the Swedish revisions of the International Classification of Diseases codes (ICD-7: 301.1, ICD-8: 296.2, ICD-9: 296B, and ICD-10: F32 and F33). The date of first hospital visit concerning depression was used as date of diagnosis for depression.
In a sensitivity analysis, we used both a broader (either ≥1 dispense of selective serotonin reuptake inhibitors (SSRIs) or clinical diagnosis of depression) and a narrower (both dispense of SSRIs and clinical diagnosis of depression) definition to assess the soundness of the main results. Information on dispense of SSRIs was derived from the Swedish Prescribed Drug Register (nationwide available since July 2005) using the Anatomical Therapeutic Chemical classification code N06AB. We also used ICD-10 codes F32.2, F32.3, F33.2 and F33.3 to identify severe depression, to assess if the role of adherence to MDP would differ for severe depression.
Covariates
We considered a range of demographic factors, lifestyle factors, anthropometric profile, and medical history as potential confounders of the studied association, including age (years, continuous), calendar year of birth (continuous), body weight (kg, continuous), height (cm, continuous), total years of education (years, continuous), smoking status (never, former or current), previous diabetes and hypertension (yes or no), as well as level of physical activity (on a 5-point scale ranging from mainly sitting as level 1 to vigorous physical activity as level 5), all collected from the questionnaires at baseline.
Statistical analysis
We calculated incidence rates of depression standardized by age using all person-time experienced by the entire cohort as the standard. Association of adherence to MDP with the risk of depression was estimated by hazard ratios (HRs) and 95% profile likelihood confidence intervals (CIs) obtained from the Cox models. The underlying time scale was attained age [27]. Adherence to MDP was analyzed both as a categorical (0-3 low, 4-5 medium, or 6-9 high) and continuous (0-9) variable. In the minimally adjusted model, we adjusted for year of birth (1942-46, 1947-51, 1952-56, or 1957-62). In the fully adjusted model, we additionally adjusted for body mass index (BMI, < 25, 25-29.9, or ≥ 30 kg/m2 calculated from weight and height) [28], years of education (0-10, 11-13, or > 13), physical activity (very low, low, moderate, high, or very high), smoking (never, former, or current), diabetes history (yes or no), hypertension history (yes or no), and total energy intake as a continuous variable (kJ/day). Since effect of diet can expect to increase cumulatively with higher age, we repeated the analyses in women younger than 50 and in women aged 50 and older. The cumulative incidence rate of depression over age by adherence to MDP was plotted using Kaplan-Meier method. Natural cubic splines were fitted to display the trend of depression risk across MDP score (0-9), adjusted for attained age, birth year, BMI, smoking, physical activity, education, diabetes, hypertension, and total energy intake.
Sensitivity and supplementary analyses
We tested the robustness of our results through a series of sensitivity analyses. To rule out the possibility of reverse causation, we excluded the first two or five years of follow-up. To address the concern for residual confounding due to a single assessment of dietary habit, we limited the follow-up time to the end of 2002 to estimate the association over the first 10 years of follow-up. We further repeated the age-specific analysis within the first 10 years and the second 10 years of follow-up respectively. We used alternative definitions for depression (broader or narrower definition, and severe depression). To address the influence of other psychiatric comorbidity, we first adjusted the analysis for history of any other psychiatric disorders (ICD-7: 300-326, ICD-8: 290-315, ICD-9: 290-319 and ICD-10: F10-99, excluding ICD codes for depression) before the end of follow-up, and then performed another analysis restricted to women without any psychiatric history before enrolment. The correlations between the nine dietary components and adherence to MDP score were calculated using Spearman’s rank correlation coefficients. To check the influence of different food components, we performed another analysis by excluding the nine components one by one from the MDP score. Finally, given the potential distinct health effects of red and white meat, we re-calculated the adherence score based only on red and processed meat [29] instead of all kinds of meat products, and also separately assessed the association of red meat with risk of depression.
The assumption of proportional hazards was assessed by examining the standardized Schoenfeld residuals [30]. All statistical tests were performed on the two-sided 5% level of significance, corresponding to a two-sided 95% CI. We did not perform any adjustment of p-values for multiplicity of statistical tests. Data management was performed using SAS software version 9. 4 (SAS institute Inc., Cary, NC, USA). Survival analyses were performed using SAS software version 9.4. The cumulative incidence rate and the age-specific analysis were performed using STATA version 14 (StataCorp LP, College Station, TX, USA). SAS codes for the Cox regression analyses are presented in the online appendix. The present study was approved by Regional Ethical Review Board in Stockholm, Sweden.