Comparison of diagnostic accuracy of models combining the renal biomarkers in predicting renal scarring in pediatric population with vesicoureteral reflux (VUR).

Introduction: Renal scarring is prominently observed in children with vesicoureteral reflux (VUR) and can lead to complicated renal outcomes. Although biopsy is the gold standard to detect renal scarring, it is an invasive procedure. There are established renal biomarkers which can help detect renal scarring. Individual biomarkers have not shown to have extensively good discriminatory ability for this.

Aim: This paper aims at combining the values of multiple biomarkers in models to detect renal scarring.

Methodology: Secondary data with the values of renal biomarkers like kidney injury molecule-1, neutrophil gelatinase-associated lipocalin (NGAL), and urinary creatinine along with the renal scarring status was considered. Logistic regression, discriminant analysis, Bayesian logistic regression, Naïve Bayes, and decision tree models were developed with these markers. The discriminatory ability of individual biomarkers along with the models was assessed using the area under the curve from ROC curve. Sensitivity, specificity, and misclassification rates were estimated and compared.

Results: NGAL was the most predominant renal biomarker in classifying the patients with renal scarring (AUC: 0.77 (0.67, 0.87); p value < 0.001). Each of the model performed better than individual biomarkers. Decision tree (AUC: 0.83 (0.74, 0.91); p value < 0.001) and Naïve Bayes model (misclassification rate = 20.2%) performed the best amongst the models.

Conclusion: Combining the values of renal biomarkers through a statistical or machine learning model to detect renal scarring is a better approach as compared to considering individual renal biomarkers.