Risk assessment of digestive cancers with the modified EAT-Lancet diet pattern: a prospective cohort study based on UK Biobank

BACKGROUND: While diet is a modifiable risk factor for digestive cancers, existing dietary scores are not optimized for site-specific risk assessment. We aimed to develop and evaluate cancer-specific dietary scores for improved risk stratification. OBJECTIVES: To modify the EAT-Lancet diet (ELD) score into cancer-specific versions and evaluate their association with the risk of seven major digestive cancers. METHODS: Based on 111,785 UK Biobank participants, we first expanded the ELD score and then derived cancer-specific scores (Score3) for seven digestive cancers using LASSO-Cox regression. Associations with cancer incidence were assessed using Cox models, Kaplan-Meier curves, and dose-response analyses. Model discrimination was evaluated using the C-index, and improvements over the original ELD score were quantified by net reclassification improvement (NRI). Subgroup analyses were conducted based on sex, smoking status, and polygenic risk score (PRS), with additive interactions between the dietary score and PRS assessed using attributable proportion (AP) and relative excess risk due to interaction (RERI), and multiplicative interactions evaluated using interaction terms in the Cox models. RESULTS: Cancer-specific Score3 showed improved discriminatory ability compared to the original ELD score, with C-index of 0.623, 0.630, 0.611, and 0.604 for esophageal, liver, rectal, and pancreatic cancers, respectively, and demonstrated positive NRI for these cancers. In multivariable-adjusted models, the lowest Score3 category was associated with a significantly higher risk for these four cancers: esophageal cancer (HR = 1.702, 95% CI: 1.202–2.410), liver cancer (HR = 1.667, 95% CI: 1.084–2.561), rectal cancer (HR = 1.662, 95% CI: 1.273–2.171), and pancreatic cancer (HR = 1.576, 95% CI: 1.201–2.069). Correspondingly, the low-score group exhibited a higher absolute risk per 10,000 person-years compared to the high-score group. A strong association was observed between low Score3 and high rectal cancer risk among individuals with high genetic predisposition (PRS); significant additive interaction was observed (AP = 0.286, RERI = 0.687). Score3 performance varied by sex and was substantially amplified among smokers. CONCLUSION: Cancer-specific modifications of the ELD score may improve risk assessment for digestive cancers, particularly esophageal, liver, rectal, and pancreatic cancers. Across digestive cancers, the associations of these modified scores also vary by sex and smoking, highlighting their potential relevance for personalized dietary guidance. The utility of the modified score is most evident in individuals with high polygenic risk for rectal cancer.