Introduction

Mental health disorders, the leading cause of avoidable suffering and premature mortality, are a growing global health crisis1. Mental health disorders affect almost 30% of individuals across their lifespan2 and were the seventh leading cause of disability-adjusted life-years (DALYs), which accounted for 125 million DALYs in 20193. Among the mental disorders, depression and anxiety are of particular concern as they ranked first and second contributors to DALYs in 20194,5. Depression and anxiety increased the risks of incident cardiovascular disease6, frailty and all-cause mortality7, and growing evidence has shown the coexistence of depression and anxiety was associated with severer consequences than each disorder alone8. Moreover, the emergence of COVID-19 pandemic has made public mental health issues eminently crucial9,10. An additional 53.2 and 76.2 million cases of depression and anxiety globally were estimated due to the COVID-19 pandemic9. Considering the substantial impairment and immense health burden imposed by depression and anxiety, evaluating them together and identifying modifiable risk factors for the primary prevention of depression and anxiety constitutes an urgent public health priority.

According to the International Society for Nutritional Psychiatry Research, nutritional medicine was suggested to be considered mainstream in psychiatric practice with the support of research, education, policy, and health promotion11,12. Due to the essential role of nutrients in the neuroendocrine system, poor diet quality resulting in unbalanced nutrient intake is an important risk factor for the development of mental disorders, and thus diet has been suggested as a promising candidate as a public health target for the prevention and treatment of these illnesses12,13. When examining the association between diet and mental disorders, dietary pattern analysis surpasses analyses of single nutrients or foods as diet patterns could reflect the numerous and multifaceted combinations of nutrient and food consumption14,15. The associations of diet patterns (e.g., Mediterranean diet16, Dietary Approaches to Stop Hypertension (DASH) diet17 and plant-based diet18,19) with the risk of depression or anxiety have also been previously reported. However, these dietary patterns mainly focus on human health. In the context of the significant negative effects on the environment due to the current food production20,21, the EAT-Lancet Commission proposed the EAT-Lancet reference diet in 2019 with the aim of increasing both planetary and human health22. The EAT-Lancet reference diet takes into account the multiple links between health, nutrition intake and the environment, and provides a set of recommendations for feeding the global population within planetary boundaries23. Overall, this dietary pattern is primarily plant based and emphasizes the intake of vegetables, fruits, whole grains and nuts. It consists of moderate amounts of seafood and poultry and considerably limits the intake of red meat, added sugar and saturated fat22. Existing observational studies have demonstrated inverse associations of the EAT-Lancet reference diet with multiple diseases and mortality24,25,26,27,28. However, evidence on the association between the EAT-Lancet reference diet and mental health is absent, and to the best of our knowledge, no studies have evaluated whether the EAT-Lancet dietary pattern is associated with the risk of develo** depression and anxiety.

In this work, we aimed to investigate the prospective associations between adherence to the EAT-Lancet reference diet and new-onset depression, anxiety and co-occurrence of depression and anxiety in a large population-based cohort of UK adults. The results show that greater adherence to the EAT-Lancet diet is associated with lower risks of incident depression, anxiety and their co-occurrence. The associations are consistent among the 3 common yet differently quantified EAT-Lancet indexes. The findings suggest that the prevention of depression and anxiety might be achieved by adhering to this sustainable dietary pattern.

Results

Baseline characteristics

A total of 180,446 participants were included in the study, among whom the mean age (standard deviation, SD) was 56.2 (8.0) years at baseline, and 83,824 (46.45%) were male (Table 1). During a median (interquartile range, IQR) follow-up of 11.62 (11.00–12.43) years, 4548, 6026 and 1262 incident cases of depression, anxiety and their co-occurrence were identified, respectively. Participants in this study obtained between 5 and 14 points on the Knuppel EAT-Lancet index, with a median (IQR) of 11 (10–11) points, between 8 and 38 points on the Stubbendorff EAT-Lancet index, with a median (IQR) of 22 (20–25) points, or obtained between −174 and 530 points on the Kesse-Guyot EAT-Lancet index, with a median (IQR) of 44 (24–67), respectively. In terms of baseline characteristics according to categories of the Knuppel EAT-Lancet index, participants with higher EAT-Lancet diet scores were more likely to be women, slightly older, nonsmokers, more physically active, and had lower body mass index (BMI) and total energy intake. Baseline characteristics according to the Stubbendorff EAT-Lancet index or Kesse-Guyot EAT-Lancet index can be seen in Supplementary Tables 1 and 2.

Table 1 Baseline characteristics of the study participants according to categories of the Knuppel EAT-Lancet index (range, 0–14 points) (N = 180,446)
Full size table

Associations of the Knuppel EAT-Lancet index (range, 0–14 points) with incident depression and anxiety

Restricted cubic splines (Fig. 1) showed that the EAT-Lancet index associated, in a linear dose-response manner, with risks of incident depression, anxiety and their co-occurrence (all P for non-linearity > 0.05). In the fully adjusted Cox regression models, the respective hazard ratios (HRs) and 95% confidence intervals (95% CIs) of incident depression, anxiety and their co-occurrence were 0.806 (0.730–0.890), 0.818 (0.751–0.892) and 0.756 (0.624–0.914) for the highest vs lowest groups of the EAT-Lancet diet index (Table 2). Each additional point of the EAT-Lancet diet score was associated with a 5.1% decreased risk of depression (HR: 0.949, 95% CI: 0.925–0.974), a 4.7% decreased risk of anxiety (HR: 0.953, 95% CI: 0.932–0.975), and a 6.3% decreased risk of co-occurrence of depression and anxiety (HR: 0.937, 95% CI: 0.892–0.985), respectively.

Fig. 1: The dose-response associations between the Knuppel, Stubbendorff and Kesse-Guyot EAT-Lancet index and risks of incident depression, anxiety and their co-occurrence.
figure 1

Solid line: Point estimation; Dotted line: 95% confidence interval; Shading: Fraction of population (Density). Multivariable Cox regression models with restricted cubic splines were adjusted for age, sex, Townsend deprivation scores, ethnicity, smoking status, alcohol intake, physical activity, hypertension, total energy intake and BMI. References are 9 for the Knuppel EAT-Lancet index, 17 for the Stubbendorff EAT-Lancet index, and 20 (the cut-off value of quintile 1) for the Kesse-Guyot EAT-Lancet index. Statistical tests were two-sided and P-value of < 0.05 was considered statistically significant. No adjustments were made for multiple comparisons. Source data are provided as a Source Data file.

Full size image
Table 2 Associations between the Knuppel EAT-Lancet index (range, 0-14) and risks of depression and anxiety
Full size table

Associations of the Stubbendorff EAT-Lancet index (range, 0–38 points) with incident depression and anxiety

Restricted cubic splines (Fig. 1) showed linear associations between EAT-Lancet index and depression, anxiety and co-occurrence of depression and anxiety (all P for non-linearity > 0.05). When comparing the highest with the lowest groups of adherence, the multivariable HR (95% CI) for incident depression was 0.711 (0.627-0.806), and the multivariable HR (95% CI) for incident anxiety was 0.765 (0.687–0.852), respectively (Table 3). When evaluating the EAT-Lancet index as a continuous variable (0–38 points), each incremental increase in points contributed to a 2.6% lower risk of incident depression (HR: 0.974, 95% CI: 0.966–0.982) and a 1.9% lower risk of incident anxiety (HR: 0.981, 95% CI: 0.974–0.988). Finally, when testing the association of the EAT-Lancet index with co-occurrence of depression and anxiety, the HRs (95% CIs) were 0.659 (0.516–0.841) for the highest adherence group and 0.971 (0.956–0.986) for 1-point increment in diet score.

Table 3 Associations between the Stubbendorff EAT-Lancet index (range, 0-38) and risks of depression and anxiety
Full size table

Associations of the Kesse-Guyot EAT-Lancet index with incident depression and anxiety

Restricted cubic splines (Fig. 1) showed linear associations between EAT-Lancet index and depression, anxiety and co-occurrence of depression and anxiety (all P for non-linearity > 0.05). After full adjustment, results from Cox regression models showed that compared with the lowest quintile (quintile 1), the HRs (95% CIs) of quintile 5 of EAT-Lancet index were 0.844 (0.768–0.928) for incident depression, 0.825 (0.759-0.896) for incident anxiety and 0.818 (0.682–0.981) for co-occurrence of depression and anxiety, respectively (Table 4). Each 100-points increase of the EAT-Lancet index was associated with a 13.6% decreased risk of depression (HR: 0.864, 95% CI: 0.795–0.938), a 15.8% decreased risk of anxiety (HR: 0.842, 95% CI: 0.784–0.905), and a 16.8% decreased risk of co-occurrence of depression and anxiety (HR: 0.832, 95% CI: 0.711–0.973).

Table 4 Associations between the Kesse-Guyot EAT-Lancet index and risks of depression and anxiety
Full size table

Predictive performance of the EAT-Lancet index

Table 5 shows that the net reclassification improvement (NRI) indexes (95% CIs) of adding the Stubbendorff EAT-lancet index to the reference model were 0.112 (0.082–0.138) for depression, 0.088 (0.062–0.116) for anxiety and 0.146 (0.058–0.202) for their co-occurrence, respectively. Likewise, adding the Kesse-Guyot EAT-Lancet index also significantly improved the risk reclassification for depression (NRI: 0.060, 95% CI: 0.026–0.092), anxiety (NRI: 0.044, 95% CI: 0.018–0.072) and their co-occurrence (NRI: 0.090, 95% CI: 0.010–0.136). Adding the Knuppel EAT-Lancet index significantly improved the risk reclassification for incident depression, with NRI (95% CI) of 0.038 (0.008–0.064). In addition, adding the Mediterranean diet index, the DASH diet index and the unhealthful plant-based diet index to the reference model brought significant improvement to the risk reclassification for incident depression, with respective NRIs (95% CIs) of 0.042 (0.012–0.060), 0.082 (0.052–0.112) and 0.112 (0.080–0.142). The corresponding NRI (95% CI) of adding the unhealthful plant-based diet index to the reference model was 0.084 (0.060–0.106) for incident anxiety and the NRIs (95% CIs) of adding the DASH diet index and the unhealthful plant-based diet index to the reference model were 0.078 (0.012–0.132) and 0.142 (0.088–0.190) for their co-occurrence, respectively.

Table 5 Net reclassification improvement for the risks of depression and anxiety associated with different diet indexes
Full size table

Sensitivity and subgroup analyses

We observed similar inverse associations between the EAT-Lancet index and incident depression, anxiety and their co-occurrence within further analyses, including excluding participants that only completed the online 24 h dietary recall questionnaire on one occasion (Supplementary Tables 3–5), using follow-up time which began at the time when the latest dietary assessment was completed (Supplementary Tables 6–8), and excluding depression, anxiety or co-occurrence cases that occurred within the first five years of follow-up (Supplementary Tables 9-11). Supplementary Tables 12–14 display the associations of individual components of the EAT-Lancet diet with risks of depression, anxiety and co-occurrence of depression and anxiety. In the Stubbendorff EAT-Lancet index, after full adjustment as in model 3 and additional adjustment for other individual food components, higher adherence to the vegetable recommendation was significantly associated with lower risks of depression, anxiety and their co-occurrence (all P for trend < 0.05). Similar associations were observed for the other two indexes. Mediation analyses showed that 22.10% (95% CI:14.60%–36.00%), 7.80% (95% CI: 3.29%–16.00%) and 18.50% (95% CI: 8.61%–60.00%) of the associations between the Knuppel EAT-Lancet index and depression, anxiety and their co-occurrence were mediated by BMI, respectively. The mediating effects of cardiovascular disease (CVD), type 2 diabetes (T2D) and hypertension were relatively weak or not significant. Similar results were observed for the Stubbendorff and the Kesse-Guyot EAT-Lancet index (Supplementary Table 15). We also explored the potential interactions between the EAT-Lancet index and age, sex, smoking status and Townsend deprivation index. We found the associations of adhering to the EAT-Lancet index with incident depression and anxiety were more pronounced in those who were more deprived, and the results were consistent among the three EAT-Lancet indexes (all P for interaction <0.05) (Supplementary Tables 16–18).

Discussion

In this prospective, population-based cohort with a median follow-up of 11.62 years, we found that greater adherence of the EAT-Lancet index was associated with lower risks of incident depression, anxiety and their co-occurrence. These associations remained consistent among the three different EAT-Lancet indexes.

The relationships between diet and depression and anxiety have gained considerable concern in recent years. Pooling all “healthy dietary patterns” from 17 studies, a 2018 meta-analysis demonstrated that adhering to a high-quality diet, regardless of type, was associated with a significantly reduced risk of depression (OR: 0.77, 95% CI: 0.69–0.84)13. More specifically, a large prospective study in UK Biobank showed that a plant-based diet rich in healthier plant foods was associated with a lower risk of incident depression (HRQ5 vs. Q1: 0.92, 95% CI: 0.85–0.99) while a plant-based diet that emphasizes less-healthy plant foods was associated with a higher risk of incident depression (HRQ5 vs. Q1: 1.15, 95% CI: 1.07–1.24)18. A cross-sectional study also suggested that a high-quality plant-based diet might be protective against depressive symptoms in Australian vegans and vegetarians19. Similar associations have been observed for anxiety29. In contrast, the prospective association of the Healthy Eating Index (HEI) with incident depressive symptoms was non-significant (HR: 0.96, 95% CI: 0.86–1.07)30 and the associations of adherence to four previously established DASH diet indices with incident depression were inconsistent and dependent on the scoring system31. Our results indicated that adhering to the EAT-Lancet diet was significantly associated with lower risks of incident depression, anxiety and their co-occurrence, and the predictive performance of the EAT-Lancet index was modestly better than or comparable to other established diet scores. It is noteworthy that all the dietary scores were calculated only based on the UK Biobank, so the results were data-driven and future studies comparing the predictive performance of different diet scores in different settings and populations are needed.

Compared with the mainstream “healthy” diet patterns, the EAT-Lancet reference diet has also been designed to serve as an anchor for integrating sustainability into national dietary recommendations of culturally diverse countries23. Using data from the European Prospective Investigation into Cancer and Nutrition (EPIC) study, Laine et al. revealed that switching from lower adherence to the EAT-Lancet reference diet to higher adherence could potentially reduce food-associated greenhouse gas emissions up to 50% and land use up to 62%, and associations of levels of dietary greenhouse gas emissions with all-cause mortality (HR: 1.13, 95% CI: 1.10–1.16) and of land use with all-cause mortality (1.18, 1.15–1.21) were observed32. In addition, higher adherence to the EAT-Lancet reference diet was associated with lower risks of multiple diseases: type 2 diabetes24,28, ischemic heart disease26 and subarachnoid stroke27, and all-cause33,34 and cause-specific mortality34. Therefore, our current findings extended previous studies by uncovering inverse associations of adherence to the EAT-Lancet reference diet with depression, anxiety and their co-occurrence, supporting that the EAT-Lancet reference diet is not only favorable for the environment health and population’s physical health, but also from a mental health aspect.

There were several approaches for translating the EAT-Lancet reference diet into the dietary scores. We adopted three common yet differently quantified EAT-Lancet indexes in the current study and the results were consistent, which indicates robust associations between adherence to the EAT-Lancet diet and incident mental outcomes. The differences among the three indices lie in the scoring criteria and food group categorizations. Compared with the binary assessments in the Knuppel EAT-Lancet index, the Kesse-Guyot EAT-Lancet index was a continuous score accounting for deviation from the cut-off value based on the Knuppel score and the Stubbendorff score used an ordinal scale, which discriminates different levels of adherence to the EAT-Lancet diet. As the emphasized individual components in all indexes, adherence to the vegetable and fruit recommendation was significantly associated with lower risks of depression, anxiety and co-occurrence of depression and anxiety. Higher intake of fish, which is emphasized in the Stubbendorff index yet limited in other two indexes, was also observed to be inversely associated with all mental outcomes, which was supported by a previous study showing that n-3 polyunsaturated fatty acids contained in fish was inversely associated with depression35. Taken together, these discrepancies in score construction may explain the differences in the predictive performances among the three EAT-Lancet indexes.

Although depression and anxiety have been considered as two distinct entities according to the diagnostic criteria, co-occurrence of depression and anxiety is relatively a common syndrome36. Among the 4548 participants who developed depression in our study, 1262 (27.7%) participants also had anxiety. Previous reports have shown that compared with individuals with single depression or anxiety, those with their coexistence had a higher risk of suicidal ideation and previous suicide attempts and had poorer responses to treatment37,38, indicating co-occurrence of depression and anxiety deserves particular concern on its management and treatment. Our study suggested that adhering to the EAT-Lancet reference diet may serve as a promising modifiable target which enhances the primary prevention for co-occurrence of depression and anxiety. In addition, we observed that the associations between adhering to the EAT-Lancet diet and incident depression and anxiety were more pronounced in those who were more deprived. Previous studies also suggested that a lower socioeconomic status was associated with greater risks of depression and anxiety39,40. Our findings suggest that the risk differences between different socioeconomic statuses might be narrowed by adhering to the EAT-Lancet diet.

The underlying mechanisms for the associations of the EAT-Lancet index with depression and anxiety have not been fully elucidated but may be explained from several aspects. First, this mainly plant-based diet pattern similar with the Mediterranean dietary pattern may reduce markers of inflammation in humans41. The inflammatory effects of a diet high in calories and saturated fat have been proposed as one mechanism that has detrimental effects on brain health, including cognitive decline, hippocampal dysfunction, and damage to the blood-brain barrier42,43. Various mental health condition, have been also linked to heightened inflammationAssessment of covariates

Covariates included sociodemographic characteristics, lifestyle factors and other potential confounding factors and were collected at baseline, except for total energy intake which was assessed by the 24-hour dietary recall questionnaire. Detailed description of the covariates can be found in Supplementary Methods 2.

Statistical analysis

For baseline characteristics, we used mean ± SD or median (IQR) to describe continuous variables and frequencies (percentages) to describe categorical variables. Kaplan-Meier curves stratified by the EAT-Lancet index are shown in Supplementary Fig. 2. Cox proportional hazards models were used to estimate HRs and 95% CIs for the associations between the baseline EAT-Lancet index and new-onset depression, anxiety and co-occurrence of depression and anxiety. The proportional hazards assumption was checked using Schoenfeld residuals. Dose-response relationships were examined by restricted cubic splines with four nodes57.

To ensure adequate numbers of participants in each group, the participants were divided into four groups for the Knuppel score ( ≤ 9, = 10, = 11, ≥ 12 points) and five groups for the Stubbendorff score ( ≤ 17, 18–20, 21–23, 24–26, and ≥ 27 points), respectively. Participants were divided into quintiles according to the Kesse-Guyot score. With the lowest group as the reference, we ran a total of three models: model 0 was unadjusted; model 1 was adjusted for age, sex, Townsend scores and ethnicity; and model 2 was adjusted for model 1 plus smoking status, alcohol intake, physical activity, hypertension, BMI and total energy intake.

To compare the predictive performance of the EAT-Lancet index and other established diet scores (the Mediterranean diet score, the DASH diet score and the plant-based diet scores) for incident mental disorders, NRI index was calculated. The NRI index could assess the proportion of cases that would be reclassified correctly by adding each diet score to the reference model58. We set the truncation time at the median follow-up time.

We performed a series of sensitivity and subgroup analyses to evaluate the robustness of our primary findings. First, we excluded those who completed the online 24 h dietary recall questionnaire on only one occasion to obtain the average intake of all dietary components as a marker of habitual intake. Second, since the four additional online dietary assessment took place later than the baseline assessment, we repeated the main analysis using the follow-up time which began at the time when the latest dietary assessment was completed. Third, considering the potential impact of reverse causality, we excluded depression, anxiety or co-occurrence cases that occurred within the first five years of follow-up. Fourth, to further explore the different roles of individual components in the EAT-Lancet index, we tested the associations between individual components and the outcomes. Fifth, mediation analyses were performed to investigate the possible pathways between adherence to EAT-Lancet diet and risks of mental disorders, and we calculated the mediation proportions (95% CIs) of BMI, CVD, T2D and hypertension. Sixth, potential interaction effects were tested by adding interaction terms between the EAT-Lancet diet indexes and covariates (age, sex, socioeconomic status and smoking status) in the model and stratified subgroup analyses were also performed.

Statistical analyses were performed using the SAS version 9.4 (SAS Institute) and R (version 4.1.1 software). All statistical tests were two-sided, and P-value of < 0.05 was considered statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.