Development and validation of cluster-specific diagnostic models for osteoporosis in chronic kidney disease patients: based on NHANES and CRDS data

Abstract

To develop and validate cluster-specific diagnostic models for osteoporosis in patients with chronic kidney disease (CKD) using routine clinical parameters. Data were obtained from the National Health and Nutrition Examination Survey. A total of 3,806 CKD patients were enrolled and randomly allocated to training (70%, n  = 2,667) and internal validation (30%, n  = 1,139) cohorts, with an additional external validation cohort from the China Renal Data System or Chinese Renal Disease Data System. K-means clustering was performed to identify distinct subgroups. Weighted univariate logistic regression, Least Absolute Shrinkage and Selection Operator and Boruta algorithm were used for feature selection. Final diagnostic models were constructed using weighted multivariable logistic regression. Model performance was assessed using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA), and net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Three distinct CKD phenotypic clusters were identified; two were selected for model development. Group A retained six variables (sex, age, body mass index, serum uric acid, serum creatinine, and fracture history) with AUCs of 0.880, 0.839, and 0.815 across training, internal, and external (CRDS) validation. Group B retained seven variables (sex, age, marital status, red blood cell count, body mass index, race, and fracture history) with AUCs of 0.824, 0.806, and 0.808. Both models demonstrated favorable calibration and clinical utility. NRI revealed divergent subgroup effects: gains in Group B (0.0059) versus losses in Group A (-0.0149). Cluster-specific diagnostic models based on weighted logistic regression exhibited robust discriminative performance for osteoporosis diagnosis in CKD patients, with heterogeneous reclassification gains across phenotypic subgroups.