Ren Fail. 2026 Dec;48(1):2700069. doi: 10.1080/0886022X.2026.2700069. Epub 2026 Jul 17.
ABSTRACT
Protein-energy wasting (PEW) is common in incident hemodialysis patients and linked to poor outcomes. The uric acid/HDL-cholesterol ratio (UHR) and intact parathyroid hormone (iPTH) relate to metabolic, inflammatory, and nutritional disturbances, but their value for predicting PEW in incident hemodialysis is unclear. This retrospective multicenter study included 863 incident hemodialysis patients. PEW was defined according to the International Society of Renal Nutrition and Metabolism criteria. UHR and iPTH were evaluated using ROC analysis, multivariable logistic regression, and ten machine learning models. Restricted cubic spline, mediation, and trajectory analyses were performed, with SHAP for interpretability. PEW was identified in 59.2% of patients. Feature selection across four machine learning approaches consistently identified UHR, iPTH, and eGFR as key predictors. Higher baseline and cumulative UHR were independently associated with a lower risk of PEW. Trajectory analysis showed that both rapidly increasing and decreasing UHR patterns with higher mean levels were linked to the lowest PEW risk. Nonlinear associations were observed, with an inverted U-shaped relationship between UHR and PEW and an S-shaped relationship between iPTH and PEW. Mediation analysis indicated that iPTH accounted for approximately 10-13% of the association between UHR and PEW. Among all models, XGBoost achieved the best performance (AUC = 0.801). A web-based Shiny tool was developed for individualized PEW risk assessment. UHR and iPTH were associated with PEW risk within specific ranges. Machine learning models integrating these markers showed favorable predictive performance and may assist individualized risk assessment in clinical practice.
PMID:42464731 | DOI:10.1080/0886022X.2026.2700069
