Objective tongue phenotyping identifies phenotypic heterogeneity in diabetic kidney disease: a dual-center clustering analysis.

Objective: To identify tongue-phenotype subtypes in patients with diabetic kidney disease (DKD) using unsupervised clustering of quantified tongue features, and to compare tongue characteristics and laboratory profiles across subtypes. Methods: We enrolled 331 patients with DKD from two hospitals in Beijing. Forty-eight continuous tongue features and 3 ordinal/categorical tongue variables were extracted. Continuous variables were z-score standardized. The optimal number of clusters was determined using the elbow method, silhouette width, and gap statistic. K-means clustering was used as the primary analysis, Ward.D2 hierarchical clustering was applied for robustness validation, and partitioning around medoids (PAM) was additionally used in sensitivity analyses. Agreement between methods was assessed by the adjusted Rand index (ARI) and maximum overlap consistency. Tongue features, 40 laboratory variables, and 7 composite indices were compared between subtypes, with Benjamini-Hochberg (BH) correction applied for multiple testing. Results: Integrating the elbow method, silhouette width, gap statistic, and interpretability, k = 2 was selected. K-means identified Cluster 1 (n = 108, 32.6%) and Cluster 2 (n = 223, 67.4%), with high concordance to hierarchical clustering (maximum overlap consistency, 92.4%; ARI, 0.715). Thirty-nine of 48 continuous tongue features remained significant after correction. Cluster 2 showed higher brightness, lower saturation, a lower coating ratio, thicker coating, and more tooth marks, whereas Cluster 1 showed darker tongue color, higher saturation, more yellow coating, and milder thickness. Coating thickness grade and tooth marks differed significantly between clusters, but fissures did not. No laboratory or composite index differed significantly after multiple imputation and BH correction. Conclusions: Objective tongue phenotyping identified two reproducible DKD subtypes. Tongue-based heterogeneity did not overlap with conventional laboratory profiles, suggesting complementary value for precision syndrome differentiation in DKD.