Metabolic Endotypes for Dialysis Risk Stratification in Chronic Kidney Disease Stages G2-G4: A Pragmatic Cluster Analysis Using Routine Laboratory Tests.

OBJECTIVE

Chronic kidney disease (CKD) affects over 850 million people globally, but traditional staging based on eGFR and albuminuria fails to capture metabolic heterogeneity and divergent outcomes in CKD stages G2-G4 (15 < eGFR < 90 mL/min/1.73m2). We aimed to identify metabolic endotypes using routine laboratory tests and link them to dialysis initiation.

METHODS

This single-center retrospective cohort study included 400 CKD stages G2-G4 patients (2022-2024) from The Central Hospital of Wuhan. Nine routine biomarkers (estimated glomerular filtration rate [eGFR], uric acid, glucose, phosphate, parathyroid hormone [PTH], albumin, hemoglobin, D-dimer, total cholesterol) underwent robust preprocessing, followed by unsupervised K-means clustering. The nine input variables were selected a priori based on their established roles in core CKD pathophysiological pathways (CKD‑MBD, malnutrition‑inflammation‑anemia, and glycolipid‑uric acid dysregulation), grounding the clustering in biological mechanisms rather than purely empirical data‑driven reduction. Optimal clusters (K=4) were confirmed via multiple statistical metrics and clinical interpretability.

RESULTS

Four distinct metabolic endotypes emerged: 1) Inflammatory-hypercoagulable (high D-dimer, low albumin); 2) Severe CKD-MBD-anemia (lowest eGFR/hemoglobin, highest PTH); 3) Metabolically favorable (highest eGFR/albumin/hemoglobin, lowest D-dimer); 4) Glycolipid-uric acid dysregulation (high glucose/uric acid/phosphate). Dialysis rates differed drastically: 59.4% (Endotype 2), 46.7% (Endotype 4), 38.7% (Endotype 1), and 6.7% (Endotype 3) (all P<0.001). Multivariable logistic regression incorporating metabolic endotype, age, gender, and eGFR yielded a corrected AUC of 0.838 (95% CI: 0.763-0.915) for predicting dialysis initiation, with a sensitivity of 94.6% and specificity of 69.4%.

CONCLUSION

This pragmatic approach, based on routinely available laboratory tests, suggests potential application value for risk stratification in clinical practice and provides a preliminary framework for future precision medicine research. However, given the single‑center retrospective design, its generalizability requires external validation in large‑scale prospective multi‑center cohorts before clinical implementation.