PubMedInternational urology and nephrology2026-11-11
Sex-specific serum metabolomic signatures of chronic kidney failure: a machine learning and interaction analysis based on the UK biobank.
Xu Xiaoting X, Wang Lihui L, Chen Qian Q, Sun Zhixia Z et al.
Whether circulating metabolites are associated with chronic kidney failure (CKF) differently in men and women remains poorly characterized, and metabolomic studies of kidney outcomes have rarely accounted for the dominant influence of renal filtration itself on the circulating metabolome. We aimed to delineate sex-specific serum metabolomic signatures of CKF under a design that separates feature selection from statistical inference and explicitly adjusts for renal function.
We analyzed nuclear magnetic resonance (NMR) metabolomic profiles from 365,422 UK Biobank participants (199,540 women, 165,882 men). Sex-stratified LightGBM models with SHAP interpretation were fitted exclusively in a random 50% discovery partition; all association estimates were derived in the independent held-out 50% validation partition using multivariable logistic regression adjusted for age, Townsend deprivation index, ethnicity, fasting status, obesity, hypertension, type 2 diabetes, and baseline serum creatinine, with Benjamini-Hochberg correction across 39 sex × metabolite interaction tests. Serum creatinine was excluded from all predictive models to avoid outcome-predictor leakage.
The metabolome provided statistically significant, but clinically modest incremental discrimination over clinical covariates (ΔAUC 0.037 [95% CI 0.031-0.043] in men and 0.026 [0.021-0.031] in women; continuous net reclassification improvement not significant). After creatinine adjustment, seven sex × metabolite interactions remained significant. The cholesteryl-ester fraction of small LDL particles was inversely associated with CKF in both sexes, but more strongly in men (OR per 1-SD 0.79 [0.77-0.81] vs 0.85 [0.83-0.87]; interaction FDR = 4.6×10⁻5), with a concordant pattern for medium LDL (FDR = 2.2×10⁻4); formal sex × spline tests confirmed steeper inverse dose-responses in men. In contrast, the apparently male-specific associations of isoleucine, alanine, and 3-hydroxybutyrate observed in unadjusted models were fully attenuated after creatinine adjustment, indicating confounding by renal filtration. Shared core metabolites were enriched for amino-acid, ketone-body, and glycolysis-related groups in both sexes (all FDR < 0.005).
Sex differences in the CKF serum metabolome are concentrated in lipoprotein cholesteryl-ester composition and remain robust after renal-function adjustment, whereas amino-acid signatures largely reflect filtration-related confounding. Adjustment for baseline renal function is indispensable in metabolomic studies of kidney outcomes.