Adherence to the EAT-Lancet diet and incident depression and anxiety

Mental disorders, especially depression and anxiety, are major contributors to global disability and mortality. The COVID-19 pandemic further increased their burden. Diet quality has been proposed as a modifiable risk factor influencing mental health via nutrient balance, inflammation, and gut–brain pathways. While prior research links Mediterranean, DASH, and plant-based dietary patterns with depression and anxiety risk, evidence is lacking for the EAT-Lancet reference diet, which integrates human and planetary health. This study investigates whether adherence to the EAT-Lancet diet is prospectively associated with incident depression, anxiety, and their co-occurrence in UK adults and evaluates these associations across three commonly used EAT-Lancet scoring systems.

Previous studies show healthy dietary patterns are associated with reduced depression risk (e.g., meta-analyses of healthy diets; protective associations for healthful plant-based diets; mixed findings for HEI and DASH depending on scoring system). Observational evidence links the EAT-Lancet diet to lower risk of cardiometabolic diseases and mortality and to reduced environmental impacts, but prior to this study there was no evidence on mental health outcomes. The EAT-Lancet diet emphasizes plant-based foods (vegetables, fruits, whole grains, nuts), moderate seafood and poultry, and low red meat, added sugars, and saturated fat. The present study addresses the evidence gap by testing the EAT-Lancet diet in relation to incident depression and anxiety.

Design: Prospective cohort analysis within the UK Biobank. Participants: 180,446 adults who completed at least one online 24 h dietary recall (Oxford WebQ). Exclusions: prior depression or anxiety at baseline; baseline use of antidepressants or anxiolytics; implausible energy intake (<500 or >3500 kcal/day for women; <800 or >4000 kcal/day for men); withdrawals. Follow-up: from baseline assessment to outcome, death, or March 23, 2021. Median follow-up was 11.62 years. Dietary assessment: Up to five 24 h recalls (one at late recruitment and four online at 3–4 month intervals, 2011–2012). Average intake across completed recalls used to reflect habitual diet. EAT-Lancet adherence: Three indices were computed:

• Knuppel index (0–14 points): binary scoring across 8 categories (whole grains; tubers/starchy veg; vegetables; fruits; dairy; protein sources; added fats; added sugars), awarding 1 point per component meeting recommendation.
• Stubbendorff index (0–38 points in original; implemented here as 0–38 with 14 components): ordinal scoring of 7 emphasized (whole grains, vegetables, fruits, fish, legumes, nuts, unsaturated oils) and 7 limited components (potatoes, dairy, beef/lamb, pork, poultry, eggs, added sugar), 0–3 points per component; higher scores indicate greater adherence.
• Kesse-Guyot index (continuous, positive/negative, here observed range −174 to 530): continuous scoring by deviation from cut-offs defined by Knuppel, capturing interindividual variability; higher scores indicate better adherence. Outcomes: Incident depression (ICD-10 F32–F33), anxiety (F40–F48), and co-occurrence (developing both during follow-up), ascertained via self-report, primary care, hospital admissions, and death registries (UK Biobank first occurrence fields). Covariates: Age, sex, ethnicity, Townsend deprivation score, smoking status, alcohol intake, physical activity, hypertension, BMI, and total energy intake. Statistical analysis: Cox proportional hazards models estimated HRs and 95% CIs comparing adherence categories (Knuppel: ≤9, =10, =11, ≥12; Stubbendorff: ≤17, 18–20, 21–23, 24–26, ≥27; Kesse-Guyot: quintiles) and per-score increments. Three models: unadjusted (Model 0); Model 1 adjusted for age, sex, Townsend, ethnicity; Model 2 additionally adjusted for smoking, alcohol, physical activity, hypertension, BMI, and energy intake. Restricted cubic splines assessed dose-response (four knots). Predictive performance was evaluated via net reclassification improvement (NRI) at median follow-up when adding each diet index to a reference model. Sensitivity analyses excluded participants with only one recall, re-anchored follow-up at the latest dietary assessment, and excluded cases in first five years to address reverse causation. Component-wise analyses assessed individual food groups. Mediation analyses estimated proportions mediated by BMI, CVD, T2D, and hypertension. Interaction tests and subgroup analyses by age, sex, smoking, and deprivation were conducted. Analyses used SAS 9.4 and R 4.1.1; two-sided P < 0.05.

• Cohort: 180,446 participants; mean age 56.2 (SD 8.0) years; 46.45% male. Incident cases: 4548 depression, 6026 anxiety, 1262 co-occurrence over median 11.62 years.
• Knuppel index (0–14): Highest vs lowest adherence HRs (95% CIs): depression 0.806 (0.730–0.890); anxiety 0.818 (0.751–0.892); co-occurrence 0.756 (0.624–0.914). Per 1-point increase: depression HR 0.949 (0.925–0.974), anxiety 0.953 (0.932–0.975), co-occurrence 0.937 (0.892–0.985). Dose-response approximately linear (all P for non-linearity > 0.05).
• Stubbendorff index (0–38): Highest vs lowest adherence HRs: depression 0.711 (0.627–0.806); anxiety 0.765 (0.687–0.852); co-occurrence 0.659 (0.516–0.841). Per 1-point increase: depression HR 0.974 (0.966–0.982), anxiety 0.981 (0.974–0.988), co-occurrence 0.971 (0.956–0.986). Linear dose-response (all P for non-linearity > 0.05).
• Kesse-Guyot index (continuous; quintiles): Quintile 5 vs 1 HRs: depression 0.844 (0.768–0.928); anxiety 0.825 (0.759–0.896); co-occurrence 0.818 (0.682–0.981). Per 100-point increase: depression HR 0.864 (0.795–0.938), anxiety 0.842 (0.784–0.905), co-occurrence 0.832 (0.711–0.973). Linear dose-response (all P for non-linearity > 0.05).
• Predictive performance (NRI when adding diet index to reference model): Stubbendorff: depression 0.112 (0.082–0.138), anxiety 0.088 (0.062–0.116), co-occurrence 0.146 (0.058–0.202); Kesse-Guyot: depression 0.060 (0.026–0.092), anxiety 0.044 (0.018–0.072), co-occurrence 0.090 (0.010–0.136); Knuppel: depression 0.038 (0.008–0.064) (anxiety/co-occurrence not significant for NRI).
• Component analyses: Higher adherence to vegetable recommendations was inversely associated with all outcomes across indices (P for trend < 0.05). Higher fish intake (emphasized in Stubbendorff) was inversely associated with outcomes.
• Mediation: For Knuppel index, estimated mediation by BMI: 22.10% (14.60%–36.00%) for depression, 7.80% (3.29%–16.00%) for anxiety, 18.50% (8.61%–60.00%) for co-occurrence. Mediation by CVD, T2D, and hypertension was weak or non-significant. Similar patterns for other indices.
• Subgroups: Associations were more pronounced among participants with higher deprivation (significant interactions across indices; all P for interaction < 0.05). Results robust in sensitivity analyses.

The study demonstrates that higher adherence to the EAT-Lancet diet is associated with lower risks of incident depression, anxiety, and their co-occurrence in a large UK cohort, with consistent linear dose-response relationships across three distinct EAT-Lancet indices. These findings fill a gap in the literature by extending the benefits of the EAT-Lancet diet—previously linked to cardiometabolic health and environmental sustainability—to mental health outcomes. Potential mechanisms include reduced systemic inflammation and beneficial modulation of the gut–brain axis through higher intake of fiber, polyphenols, and unsaturated fats. Component analyses highlight the importance of vegetables and fish. The stronger associations among more socioeconomically deprived participants suggest that dietary adherence could help reduce mental health disparities. Predictive performance analyses show that EAT-Lancet indices, particularly the Stubbendorff and Kesse-Guyot versions, modestly improve risk reclassification beyond standard risk factors and perform comparably to or better than other diet scores in this dataset.

In this prospective UK Biobank study, greater adherence to the EAT-Lancet reference diet—captured by three different scoring methods—was associated with reduced risks of incident depression, anxiety, and their co-occurrence. Promoting this sustainable, plant-forward dietary pattern may yield mental health benefits in addition to environmental and physical health gains. Future research should validate these findings in more diverse populations, compare diet scores across settings, elucidate biological mechanisms (e.g., inflammation and microbiome pathways), and assess intervention effectiveness and equity impacts.

• Outcome misclassification and delayed diagnoses are possible despite multiple data sources; symptom presence could alter diet.
• Selection bias: over half of the cohort lacked dietary recalls and were excluded; although differences were small, generalizability may be affected.
• Diet is time-varying; repeated measures were limited to up to five recalls over roughly a year, which may not capture long-term changes.
• Adaptation of the 42-point Stubbendorff score to UK Biobank data was imperfect due to limited detail on unsaturated oils and red meat subtypes; however, impact on discrimination appeared minimal.
• UK Biobank participants are not fully representative of the general population, and the analytic sample was >95% White, limiting generalizability.