A clinically applicable CKD diagnostic model derived from sound touch viscosity ultrasound and LASSO regression

Background chronic kidney disease (CKD) remains a global health challenge with limitations in current diagnostic methods, including the invasiveness of biopsies and variability of estimated glomerular filtration rate (eGFR). This study aimed to develop a noninvasive diagnostic model integrating ultrasound viscoelasticity parameters to address these gaps. Methods A prospective cohort of 228 participants underwent standardized renal ultrasound with viscoelastic imaging (Mindray Resona A20) to assess viscoelastic parameters and structural metrics. Key predictors were selected through LASSO regression, and a logistic regression diagnostic model was constructed. Model performance was comprehensively evaluated by analyzing discriminative ability (AUC, sensitivity/specificity), calibration (Brier score, calibration curves), and clinical utility (nomogram development, risk stratification and stage-specific decision curve analysis). Multiclass analysis was implemented to evaluate stage-specific performance (Class 1: normal; Class 2: G1-3; Class 3: G4-5) using one-vs-rest ROC methodology. All statistical analyses incorporated 1000 bootstrap iterations for robust variance estimation. Results The diagnostic model demonstrated superior accuracy with an AUC of 0.932 (95% CI 0.908-0.956) in validation sets. Pathological analysis revealed that viscosity values were significantly elevated in CKD patients compared to controls (1.99 vs. 1.64 Pa·s, P < 0.001), while elasticity and shear wave velocity showed increases of 12.7%-13.2% and 5.3% respectively ( P < 0.001). For clinical implementation, the model incorporated a visual nomogram that converted scores ranging from 0 to 160 points into CKD probability estimates between 0.1 and 0.9, with an optimal cutoff value of 0.383 providing balanced sensitivity of 88.4% and specificity of 87.8%. Decision curve analysis confirmed clinical utility across probability thresholds of 20%-80%, with peak net benefit at 40% threshold probability. Multiclass analysis revealed stage-dependent performance: Class 3 showed the highest discrimination (AUC = 0.918), followed by Class 1 (AUC = 0.884) and Class 2 (AUC = 0.774), with significant inter-stage differences (DeLong’s test P < 0.001). Conclusion This study establishes a novel “function-structure” integrated diagnostic paradigm for CKD, combining the accuracy of ultrasound parameters with unique structural insights. The model’s noninvasive nature and stability under physiological variability make it particularly valuable for early detection and longitudinal monitoring.