Abstract / Summary
Diabetic kidney disease (DKD) has a heterogeneous clinical course. A refined classification of DKD phenotypes could help identify patients at increased risk of kidney disease progression and guide individualized clinical management. This retrospective cohort included 6,667 adults with type 2 diabetes and DKD. Phenotypes were derived by outcome-blind partitioning around medoids using 11 routinely available baseline variables. Reproducibility was evaluated using repeated subsampling and consensus clustering across multiply imputed datasets. Cox models assessed kidney outcomes, and linear mixed-effects models characterized longitudinal estimated glomerular filtration rate (eGFR) trajectories. Phenotypic and prognostic heterogeneity within specific combinations of glomerular filtration rate and albuminuria categories defined by Kidney Disease: Improving Global Outcomes (KDIGO) was also examined. Three phenotypes were identified: metabolic–preserved kidney function (P1; n = 3,797), older, leaner, lower-eGFR (P2; n = 1,947), and albuminuric renal impairment (P3; n = 923). Among 2,651 patients included in the time-to-event analysis, 546 experienced the composite kidney outcome. Compared with P1, the adjusted hazard ratios were 1.89 (95% CI, 1.50–2.37) for P2 and 8.12 (95% CI, 6.54–10.07) for P3. Adjusted annual eGFR slopes were −4.94, −3.80, and −9.14 mL/min/1.73 m 2 per year for P1, P2, and P3, respectively. All 15 KDIGO category combinations contained more than one phenotype. After stratification by KDIGO category combination, the adjusted hazard ratios were 1.35 (95% CI, 1.03–1.79) for P2 and 2.35 (95% CI, 1.70–3.26) for P3. Outcome-blind clustering identified three DKD phenotypes with distinct kidney outcome risks and longitudinal eGFR trajectories. Multidimensional clinical phenotyping may complement KDIGO classification.