Untitled

ObjectiveTo examine associations between objectively quantified tongue features and the Controlling Nutritional Status (CONUT) score in patients with diabetic kidney disease (DKD), and to assess the influence of renal function and glycaemic status on these associations.MethodsThis dual-centre cross-sectional study included 392 patients with DKD. Fifty-one tongue features were extracted using the YZAI-02 AI tongue imaging system. CONUT was calculated from serum albumin, lymphocyte count, and total cholesterol after multiple imputation by chained equations (m = 20). Rubin-pooled partial Spearman correlations, adjusted single-feature and joint multivariable linear regression, sensitivity analyses, and ordinal logistic regression were performed. Covariates were selected a priori and included age, sex, study centre, estimated glomerular filtration rate (eGFR), and HbA1c. Exploratory stratified and interaction analyses were conducted according to eGFR and sex.ResultsThe median CONUT score was 3 (interquartile range: 2–5), and 304 patients (77.6%) had CONUT-defined malnutrition. Five of 51 tongue features reached nominal significance in partial-correlation screening, although none survived false discovery rate correction across all 51 comparisons. After collinearity pruning, five representative features met the exploratory false discovery rate threshold in adjusted single-feature models. In the joint multivariable model, higher edge tongue saturation (β = 0.269 per standard deviation, 95% CI: −0.023 to 0.560, p = 0.071) and lower middle tongue brightness (β = −0.234, 95% CI: −0.495 to 0.026, p = 0.078) showed borderline associations with higher CONUT scores. Adding eGFR increased the coefficient for edge tongue saturation from 0.131 to 0.289, consistent with statistical suppression, whereas the coefficient for middle tongue brightness remained stable. Additional proteinuria adjustment produced coefficient changes below 15%, and both estimates remained imprecise. In ordinal logistic regression, edge tongue saturation showed a directionally consistent borderline association (OR = 1.29, 95% CI: 0.96–1.71, p = 0.086). No interaction remained significant after false discovery rate correction.ConclusionCONUT-defined malnutrition was common in DKD. Middle tongue brightness and edge tongue saturation showed modest, borderline associations with nutritional risk, suggesting that AI-assisted tongue phenotyping warrants further evaluation as a complement to nutritional surveillance in DKD.