Selection and Validation of Novel Biomarkers for ntOPN-Based Models for Diabetic Kidney Disease in Patients With Diabetes Mellitus.
We previously found that urinary n-terminal osteopontin (ntOPN) performed well for predicting diabetic kidney disease (DKD). This study is aimed at screening potential biomarkers for improving ntOPN-based models in DKD detection and prediction.
We performed a cross-sectional and then prospective cohort study. The novel biomarkers for DKD development were selected by the SOMAscan platform. The selected biomarkers were further validated by the SHapley Additive exPlanations (SHAP) algorithm, Pearson correlation, and logistic regression. The ntOPN-based models for DKD prediction were established, evaluated, and utilized by machine learning.
The baseline growth differentiation factor 15 (GDF15) was selected by SOMAscan assays, and urinary GDF15 was validated as an independent predictor for DKD occurrence (adjusted OR 1.43, 95% CI 1.20-1.75) and progression (adjusted OR 1.39, 95% CI 1.15-1.75) by multivariate logistic regression. The receiver operating characteristic (ROC) analysis showed that the multibiomarker panel consisting of urinary ntOPN-to-creatinine ratio (UntOCR) and urinary GDF15-to-creatinine ratio (UGCR) had stronger abilities in forecasting the 2-year risk of DKD occurrence (AUC 0.838 vs. 0.818) and DKD progression (AUC 0.867 vs. 0.834) than the combination of estimated glomerular filtration rate (eGFRcr-cys) and urinary albumin-to-creatinine ratio (UACR). A nomogram was further built with a high C-index (0.8433).
Compared with eGFRcr-cys combined with UACR, the models based on urinary ntOPN and GDF15 could provide more accurate tools for DKD prediction. Our attempt might provide a feasible approach for searching promising biomarkers for clinical applications.
We performed a cross-sectional and then prospective cohort study. The novel biomarkers for DKD development were selected by the SOMAscan platform. The selected biomarkers were further validated by the SHapley Additive exPlanations (SHAP) algorithm, Pearson correlation, and logistic regression. The ntOPN-based models for DKD prediction were established, evaluated, and utilized by machine learning.
The baseline growth differentiation factor 15 (GDF15) was selected by SOMAscan assays, and urinary GDF15 was validated as an independent predictor for DKD occurrence (adjusted OR 1.43, 95% CI 1.20-1.75) and progression (adjusted OR 1.39, 95% CI 1.15-1.75) by multivariate logistic regression. The receiver operating characteristic (ROC) analysis showed that the multibiomarker panel consisting of urinary ntOPN-to-creatinine ratio (UntOCR) and urinary GDF15-to-creatinine ratio (UGCR) had stronger abilities in forecasting the 2-year risk of DKD occurrence (AUC 0.838 vs. 0.818) and DKD progression (AUC 0.867 vs. 0.834) than the combination of estimated glomerular filtration rate (eGFRcr-cys) and urinary albumin-to-creatinine ratio (UACR). A nomogram was further built with a high C-index (0.8433).
Compared with eGFRcr-cys combined with UACR, the models based on urinary ntOPN and GDF15 could provide more accurate tools for DKD prediction. Our attempt might provide a feasible approach for searching promising biomarkers for clinical applications.