Neural Network and Calculator for Predicting Breakthrough Febrile Urinary Tract Infections in Children with Primary Vesicoureteral Reflux Open Access

Abstract

Background: Factors that influence the decision to surgically correct VUR include risk of developing new renal parenchymal scarring, further risk of breakthrough fUTI, and the likelihood of spontaneous VUR resolution. Improved identification of children at risk for breakthrough fUTI could help make decisions regarding whether or not to surgically correct VUR. We constructed three models to predict breakthrough fUTI and evaluated the accuracy of these models.

Methods and Materials: Medical records of 384 children diagnosed with primary VUR in whom detailed voiding cystourethrogram (VCUG) and clinical data were documented between 1984 and 2010 were reviewed. The variables involved in our analyses included age, gender, percentage of PBC at VUR onset, VUR grade (0-2, 3, 4-5), laterality, history of BBD, number of UTIs prior to VUR diagnosis (≥2 vs. <2), history of fUTI, VUR onset (filling or voiding) and dilating VUR (0-2, 3-5). The data was randomized into a training set of 288 for model creation and a validation (testing) set of 96, following the 75/25-splitting rule. The data was modeled with logistic regression, neural network, and random forest. Receiver operating characteristic (ROC) area was utilized to assess the performance of the model.

Results: A total of 384 patients with primary reflux were recruited in the study, with 64 male patients (16.67%) and 320 female patients (83.33%). The number of patients developing the outcome of breakthrough fUTI was 128 (33.33%). Gender, percentage of PBC at VUR onset, VUR grade, bilateral VUR, BBD, history of fUTI were significantly associated with breakthrough fUTI. The model that best fit the data and had the highest discrimination ability was a one-hidden node neural network model, with an AUC of 0.756.

Conclusion: Our neural network model, using multiple variables, predicts breakthrough fUTI on an individual basis with an AUC of 76%. Our prognostic calculator based on the neural network model will provide a useful tool for users to conveniently get a prediction and a probability of developing breakthrough fUTI. Such prognostic information can assist in clinical decision-making and provide useful information in reflux management.

Table of Contents

CONTENTS

1. INTRODUCTION

2. METHODS AND MATERIALS

3. RESULTS

4. CONCLUSION

5. DISCUSSION

6. REFERENCE

7. TABLES AND FIGURES

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Rollins School of Public Health
Department	Biostatistics
Subfield / Discipline	Biostatistics - MPH & MSPH
Degree	M.S.P.H.
Submission	Master's Thesis
Language	English
Research Field	Biology, Biostatistics Health Sciences, General
Keyword	English Prediction Breakthrough febrile urinary tract infection Neural network
Committee Chair / Thesis Advisor	Leong, Traci, Emory University
Committee Members	Ko, Yi-an, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Neural Network and Calculator for Predicting Breakthrough Febrile Urinary Tract Infections in Children with Primary Vesicoureteral Reflux ()	2020-04-14 16:31:38 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions