The EAT–Lancet planetary health diet and risk of incident chronic kidney disease

Data source and study population

The UK Biobank is a large-scale longitudinal study comprising about 500 000 participants aged 40 to 69 years from England, Scotland, and Wales.¹⁶ At enrolment, participants completed a touch-screen questionnaire on demographic characteristics, lifestyle, and health, and underwent physical examinations and biological sample collection. Representative subsets of participants provided additional data for metabolic and proteomic analyses.

The study included participants who completed at least 1 online 24-hour dietary recall questionnaire. We excluded participants who had extreme energy intake (male > 4200 or < 600 kcal/d; female > 3600 or < 500 kcal/d); missing baseline CKD status data; and prevalent CKD at baseline (defined as self-reported CKD, estimated glomerular filtration rate [eGFR] < 60 mL/min/1.73 m², urine albumin-to-creatinine ratio [UACR] ≥ 30 mg/g, or CKD diagnosis before baseline assessment).

Dietary assessment and EAT–Lancet diet index

Dietary information was collected using the Oxford WebQ Questionnaire, a Web-based 24-hour dietary recall questionnaire.¹⁷ Food item definitions are provided in Appendix 1, Supplementary Table S1, available at www.cmaj.ca/lookup/doi/10.1503/cmaj.250457/tab-related-content.

Multiple scoring methods have been proposed to quantify adherence to the EAT–Lancet Commission dietary recommendations, though no consensus exists. To ensure robustness in examining associations with incident CKD, we employed 4 established EAT–Lancet diet scores (Box 1).

Higher scores in all 4 methods reflect greater adherence to the EAT–Lancet diet. Among these, the Kesse-Guyot score offers superior ability to capture individual dietary variation by quantifying continuous deviations from recommended cut-off values. ²¹ This method improves discriminative power in assessing diet adherence²² and is especially effective for detecting subtle, diet-related differences in metabolic and proteomic profiles. Therefore, unless otherwise stated, we considered the Kesse-Guyot score to be the primary measure of EAT–Lancet diet adherence.

We also calculated other dietary indices — including DASH,⁵ aMed,⁶ AHEI-2010,⁷ and hPDI⁸ — and summarized them in Appendix 1, Supplementary Table S5. Dietary components across patterns are compared in Appendix 1, Supplementary Table S6.

Outcome assessment

The study outcome was incident CKD, defined using the International Classification of Diseases, Tenth Revision (ICD-10) code N18, consistent with previous epidemiologic studies.^23–25 Case identification incorporated data from primary care records, hospital admissions, and death registries to enhance completeness.

We calculated follow-up time from the date of baseline assessment to the date of incident CKD, death, or the end of follow-up (Sept. 30, 2021, for England; July 31, 2021, for Scotland; and Feb. 28, 2018, for Wales), whichever came first.

Statistical analysis

Baseline characteristics are presented as means (standard deviation [SD]) for continuous variables and proportions for categorical variables. We assessed the association between the EAT–Lancet index (per 1-SD increment and by quartiles) and the risk of incident CKD using Cox proportional hazards models, with results expressed as hazard ratios (HRs) and 95% confidence intervals (CIs). A directed acyclic graph depicting the assumed relationships among the EAT–Lancet diet index, CKD risk, and all covariates is provided in Appendix 2, Supplementary Figure S1, available at www.cmaj.ca/lookup/doi/10.1503/cmaj.250457/tab-related-content. We performed sensitivity analyses and conducted subgroup analyses to evaluate potential effect modifiers, including genetic factors (genetic risk score and single-nucleotide polymorphisms for CKD) and environmental factors (residential green and blue spaces). We evaluated the credibility of subgroup effect modifications using the Instrument to assess the Credibility of Effect Modification Analyses (ICEMAN) tool.²⁶

We constructed metabolic and proteomic signatures using a 2-step analysis: multivariable-adjusted linear regressions followed by the least absolute shrinkage and selection operator (LASSO) model. We calculated signature scores using a weighted sum of identified metabolites and proteins. We derived pathway scores from metabolite categories and protein enrichment. Mediation analysis estimated the proportion of the association between the EAT–Lancet index and incident CKD risk mediated by these signatures and pathways.

We conducted all analyses using R version 4.1.1, with a 2-tailed p < 0.05 considered statistically significant. Additional methodologic details regarding dietary assessment and EAT–Lancet diet index, metabolomic and proteomic profiling, genetic and environmental assessment, covariate assessment, and statistical approaches are provided in Appendix 3, available at www.cmaj.ca/lookup/doi/10.1503/cmaj.250457/tab-related-content.

Ethics approval

The study received ethics approval from the North-West Multi-Centre Research Ethics Committee (11/NW/0382), and all participants provided written informed consent.

Study participants and population characteristics

The enrolment period for participants spanned from December 2006 to October 2010. Among the 179 508 study participants (Figure 1), the mean age was 55.9 (SD 7.9) years, 54.6% were female, and 96.0% were White. Those who developed incident CKD were older and less likely to be never-smokers. They also exhibited higher body mass index; lower physical activity, alcohol consumption, eGFR, and EAT–Lancet index scores; and a higher prevalence of hypertension, hypercholesterolemia, and diabetes (Table 1). Participant characteristics were consistent across all analytic samples, as well as between individuals included and those not included in the metabolomics and proteomics analyses (Appendix 1, Supplementary Tables S7 to 9).

Association between EAT–Lancet Diet adherence and incident CKD

During a median follow-up of 12.1 (interquartile range 11.5 to 12.9) years, 4819 (2.7%) participants developed CKD. The median values for the Yi-Xiang, Knuppel, Stubbendorff, and Kesse-Guyot scores were 50, 9, 21, and 22.9, respectively.

In fully adjusted Cox models, higher scores for each diet index (per 1-SD increase) were associated with a lower risk of incident CKD: Yi-Xiang (adjusted HR 0.94, 95% CI 0.91 to 0.97), Knuppel (adjusted HR 0.94, 95% CI 0.92 to 0.97), Stubbendorff (adjusted HR 0.91, 95% CI 0.88 to 0.94), and Kesse-Guyot (adjusted HR 0.92, 95% CI 0.90 to 0.95) (Table 2). When assessed by quartiles, the Kesse-Guyot score showed a dose–response relationship, with adjusted HRs of 0.90 (95% CI 0.84 to 0.98) for Q2, 0.86 (95% CI 0.79 to 0.93) for Q3, and 0.82 (95% CI 0.76 to 0.90) for Q4 compared with Q1. We observed similar trends for the other scores. Among the 4, Stubbendorff and Kesse-Guyot scores exhibited stronger inverse associations, with a significantly reduced risk already apparent in Q2 (Table 2). Adjusted HRs (95% CIs) for all variables in the multivariable Cox regression models are provided in Appendix 1, Supplementary Tables S10 to 17.

Sensitivity analyses — including further adjustment for a variety of additional food groups and environmental factors, exclusion of early CKD cases or participants with only 1 dietary assessment, use of the last available diet record as the baseline, or complete-case analysis — did not materially alter the results (Appendix 1, Supplementary Table S18). Removing individual EAT–Lancet index components had minimal effect (Appendix 1, Supplementary Table S18). Additionally, in a subsample with repeated eGFR measurements in 2012 and 2013, the inverse association for EAT–Lancet index was consistent when using a biochemical CKD definition (eGFR < 60 mL/min/1.73 m²), supporting the robustness of our primary findings (Appendix 1, Supplementary Table S18).

Exploratory analyses also showed inverse associations between higher diet scores (per 1-SD increase) and CKD risk for DASH (adjusted HR 0.91, 95% CI 0.88 to 0.93), aMed (adjusted HR 0.92, 95% CI 0.90 to 0.95), hPDI (adjusted HR 0.93, 95% CI 0.90 to 0.96), and AHEI-2010 (adjusted HR 0.96, 95% CI 0.93 to 0.99) (Appendix 1, Supplementary Table S19).

Subgroup analysis

In subgroup analyses of genetic factors, the CKD genetic risk score did not significantly modify the association of incident CKD with either version of the EAT–Lancet index (Kesse-Guyot or Stubbendorff score) or with other dietary patterns (DASH, aMed, AHEI-2010, or hPDI) (Figure 2 and Appendix 2, Supplementary Figure S2). However, the single-nucleotide polymorphism rs2010352-G showed a significant interaction with the EAT–Lancet index on CKD risk (p for interaction < 0.001) (Figure 2 and Appendix 1, Supplementary Table S20), which was rated as moderately credible by the ICEMAN tool (Appendix 4, available at www.cmaj.ca/lookup/doi/10.1503/cmaj.250457/tab-related-content).

For environmental factors, stronger inverse associations between the EAT–Lancet index (per 1-SD increase in Kesse-Guyot score) and CKD were observed among participants with lower residential green space (1000-m buffer: p for interaction = 0.008; 300-m buffer: p for interaction = 0.09) and lower blue space at 300 m (p for interaction = 0.1) (Figure 2). We noted similar trends for the Stubbendorff-based score (Appendix 2, Supplementary Figure S2). The interaction with residential green space (1000-m buffer) was considered moderately credible per ICEMAN (Appendix 4). No environmental factors significantly modified associations between CKD risk and the other dietary patterns (Figure 2).

We observed no significant effect modification for age, sex, race, Townsend deprivation index, physical activity, or history of diabetes on the association between the EAT–Lancet index and CKD (all p for interaction > 0.05; Appendix 2, Supplementary Figures S2 and S3).

Metabolic and proteomic signatures of the EAT–Lancet planetary health diet

Linear regression followed by Bonferroni correction identified 146 metabolites significantly associated with the EAT–Lancet index (Kesse-Guyot score) (Appendix 1, Supplementary Table S21). LASSO regression refined these to 122 key metabolites, primarily comprising lipoprotein subclasses and fatty acids (Appendix 1, Supplementary Table S22; Appendix 2, Supplementary Figure S4). Descriptive statistics of the resulting metabolic signature score are detailed in Appendix 1, Supplementary Table S23. This score showed a significant positive correlation with the EAT–Lancet index (Spearman r = 0.2, p < 0.001; Appendix 2, Supplementary Figure S5).

Similarly, 420 proteins were significantly associated with the index after Bonferroni correction, with LASSO regression narrowing these to 143 key plasma proteins (Appendix 1, Supplementary Tables S24 and S25; Appendix 2, Supplementary Figure S4). Kyoto Encyclopedia of Genes and Genomes enrichment analysis indicated their involvement in cytokine–cytokine receptor interactions, phagosome formation, cell adhesion molecules, and complement and coagulation cascades (Appendix 1, Supplementary Table S26). Detailed descriptive statistics of the resulting proteomic signature score are provided in Appendix 1, Supplementary Table S23. The proteomic signature score was positively correlated with the EAT–Lancet index (Spearman r = 0.3, p < 0.001; Appendix 2, Supplementary Figure S5).

Metabolomic and proteomic profiles were also established for the DASH, aMed, hPDI, and AHEI-2010 diets (Appendix 1, Tables S27 to 34). Notably, 68 metabolites and 45 proteins were shared between the EAT–Lancet index and all other dietary patterns (Appendix 2, Supplementary Figure S6).

Association of metabolic and proteomic signature scores of the EAT–Lancet index with incident CKD risk

In the metabolic and proteomic cohorts, 2716 (2.8%) and 548 (2.9%) participants developed CKD, respectively. As shown in Figure 3A and 3B, both the metabolic and proteomic signature scores of the EAT–Lancet index exhibited significant inverse linear associations with CKD risk. Each 1-SD increase in the metabolic signature score was associated with an 11% lower risk of CKD (adjusted HR 0.89, 95% CI 0.85 to 0.93) (Figure 3A). Similarly, each 1-SD increase in the proteomic signature score was associated with a 20% risk reduction (adjusted HR 0.80, 95% CI 0.73 to 0.89) (Figure 3B). Furthermore, participants with high metabolic or proteomic signature scores showed a lower cumulative incidence of CKD than those with high EAT–Lancet index scores (Figure 3C and 3D).

Mediation analysis

The metabolic signature score significantly mediated 18.0% of the total effect of the EAT–Lancet index on CKD risk (p < 0.001; Figure 4A). Pathway-specific analysis identified 3 major mediating metabolic categories: fatty acids (12.0%), inflammation (8.6%), and fluid balance (2.9%) (Figure 4B). Key mediators included the degree of unsaturation (the number of carbon–carbon double bonds in fatty acids), glycoprotein acetyls, and docosahexaenoic acid (DHA) (Appendix 1, Supplementary Table S35; Appendix 2, Supplementary Figure S7).

The proteomic signature score mediated a more substantial proportion of the association, accounting for 27.2% of the total effect (p < 0.001; Figure 4C). Key mediating pathways included cytokine–cytokine receptor interaction (15.1%), efferocytosis (16.5%), and lysosome function (6.6%) (Figure 4D). Major protein mediators were hepatitis A virus cellular receptor 1 (HAVCR1), interleukin-18 receptor 1 (IL18R1), and pro-transforming growth factor α (Appendix 1, Supplementary Table S36; Appendix 2, Supplementary Figure S7).