Web-based multi-variable computational models predicting early spontaneous resolution in children with vesicoureteral reflux.

PURPOSE: Multi-variable computational models have been used in clinical practice to help assist decision making, including predicting clinical outcomes for vesicoureteral reflux (VUR). We reviewed predictive factors of spontaneous resolution, including distal ureteral diameter ratio (UDR). We then updated online computational models predicting spontaneous VUR resolution at 1 and 2 years to assist management of VUR on a more individualized basis. METHODS: We reviewed the records of 828 patients with VUR, including 616 females and 212 males. In addition to UDR, other prognostic clinical variables were evaluated. Specifically, the outcome of spontaneous resolution versus no resolution was analyzed using available data for each predictor set at 1 and 2 years after voiding cystourethrogram (VCUG) using logistic models with various combinations of clinical variables. Not all patients had complete data for all variables, and analyses were performed using the subset of patients with complete data for each model. ROC curves with 10-fold cross-validation were constructed to assess predictive performance. RESULTS: All single model predictors including reflux grade, volume at onset of reflux, and UDR demonstrated better ability in predicting VUR resolution compared to covariate models alone. The models were deployed using a web application created with R Shiny, an open-source package for the programming language R that is used to build interactive web applications. The app allows for patient-specific variables to be entered and produces predictions of outcomes at 1 and 2 years post VCUG. CONCLUSIONS: These computational models permit the input of multiple factors for a more individualized approach to predicting spontaneous resolution within 1 and 2 years of VCUG.