Digital Phenomics and Dual-Fluid Metabolomics for Precision Stratification of Diabetic Kidney Disease: A Multicenter Cohort Study

We conducted a prospective-retrospective two-center cohort study involving 364 participants. The Discovery Cohort (n=282) enrolled from two centers consisted of healthy controls (n=80), patients with type 2 diabetes mellitus (T2DM) without kidney disease (n=102), and patients with confirmed DKD (n=100). An independent External Validation Cohort (n=82) was established to assess model generalizability. We developed a standardized 'Digital Urine' analysis pipeline using computer vision algorithms to quantify chromatic and foam stability phenotypes. Concurrently, dual-fluid metabolomics (untargeted LC-MS/MS and targeted 1H-NMR) was performed on serum and urine to map metabolic flux. The primary outcome was the construction and validation of a multi-modal disease classifier. The secondary outcome was the identification of metabolite flux ratios reflecting renal tubular transporter function.