Association and predictive value of the C-reactive protein-triglyceride-glucose index in chronic kidney disease

Abstract Recent evidence has suggested a positive association between the C-reactive protein-triglyceride-glucose index (CTI) and chronic kidney disease (CKD), but its generalizability and prognostic utility remain to be fully established. In this study, we examined the CTI and CKD association in a nationally representative US cohort and further evaluated its prognostic value for mortality using machine learning, with external validation in an independent Chinese population. We analyzed data from the National Health and Nutrition Examination Survey (NHANES) from 2001 to 2010. Logistic regression models were used to assess the association between CTI (as a continuous variable and in quartiles) and CKD. Cox regression models were applied to assess the relationship between CTI and all-cause mortality among patients with CKD. Nonlinear relationships between CTI and all-cause mortality were evaluated using restricted cubic splines (RCS). Furthermore, we further adopted machine learning to evaluate the prognostic value of CTI, and performed mediation analysis to investigate the roles of diabetes and body mass index(BMI) in this association. CTI showed significant nonlinear associations with both CKD prevalence and all-cause mortality, with thresholds at 9.868 and 10.170, respectively. In cross-sectional analyses, these associations were stronger in younger individuals, males, participants with diabetes and obesity. In longitudinal analyses, each one-unit increase in CTI was associated with a 15% higher all-cause mortality (HR = 1.15; 95%CI:1.02–1.30), but not cardiovascular death. Internally, GBM and XGBoost outperformed other models, but their external performance degraded. CoxBoost, showing the highest external validity, was ultimately chosen as the preferred model for its better generalizability. SHAP analysis identified CTI as a modest yet independent predictor, ranking above diabetes, BMI, and total cholesterol. Exploratory mediation analysis showed that diabetes accounted for only 3.1% of the association between CTI and mortality, whereas the indirect effect of BMI was opposite in direction to the total effect. CTI was nonlinearly and threshold-dependently associated with CKD risk and all-cause mortality. Machine learning models incorporating CTI showed predictive performance, supporting its potential value for CKD risk stratification and prognostic assessment.