82 / 492
08:54 - 08:57
S4-6
(PP)
COMPUTER MODEL PREDICTING BREAKTHROUGH FEBRILE UTI IN CHILDREN
WITH PRIMARY VESICOURETERAL REFLUX (VUR)
Siobhan E. ALEXANDER, Moshe WALD, Angela M. ARLEN and Christopher S. COOPER
University of Iowa, Department of Urology, Iowa City, USA
PURPOSE
We developed and investigated the accuracy of a new multi-variable computational model for predicting breakthrough
fUTIs in children with VUR.
MATERIAL AND METHODS
Children with primary VUR and detailed clinical and voiding cystourethrogram (VCUG) data were identified. Patient age,
gender, VCUG findings including grade, laterality, bladder volume at onset of VUR, UTI history, and presence of
bladder/bowel dysfunction were assessed to determine likelihood of breakthrough fUTI. Median follow-up was 24 months
(interquartile range 12 to 52 months). The dataset was randomized into a training set of 288 and a separate
representational cross-validation set of 96. Various model types and architectures were investigated using neUROn++, a
set of C++ programs.
RESULTS
Two hundred fifty children (208 girls, 47 boys) diagnosed with primary VUR at a mean age of 3.1 years (± 2.6) met all
inclusion criteria; 384 VCUGs were analyzed. Sixty-eight children (26.7%) experienced 90 breakthrough fUTI events. A
2-hidden node neural network model had the best fit with an ROC area of 0.755 for predicting breakthrough fUTI. A
prognostic calculator using this model can be deployed for availability on the internet, allowing input variables to be
entered to calculate the odds of developing a breakthrough fUTI.
CONCLUSIONS
This is the first computational model using multiple variables including bladder volume at onset of VUR. It provides
improved individualized prediction of children at risk for breakthrough fUTI. A web-based prognostic calculator based on
this model will provide a useful tool for assessing the risk of breakthrough fUTI in children with primary VUR.