Targeted Workup After Initial Febrile Urinary Tract Infection: Using a Novel Machine Learning Model to Identify Children Most Likely to Benefit From Voiding Cystourethrogram

Advanced Analytics Group of Pediatric Urology and ORC Personalized Medicine Group

PMID: 30810465
DOI: 10.1097/JU.0000000000000186

Abstract

Purpose: Significant debate persists regarding the appropriate workup in children with an initial urinary tract infection. Greatly preferable to all or none approaches in the current guideline would be a model to identify children at highest risk for a recurrent urinary tract infection plus vesicoureteral reflux to allow for targeted voiding cystourethrogram while children at low risk could be observed. We sought to develop a model to predict the probability of recurrent urinary tract infection associated vesicoureteral reflux in children after an initial urinary tract infection.

Materials and methods: We included subjects from the RIVUR (Randomized Intervention for Children with Vesico-Ureteral Reflux) and CUTIE (Careful Urinary Tract Infection Evaluation) trials in our study, excluding the prophylaxis treatment arm of the RIVUR. The main outcome was defined as recurrent urinary tract infection associated vesicoureteral reflux. Missing data were imputed using optimal tree imputation. Data were split into training, validation and testing sets. Machine learning algorithm hyperparameters were tuned by the validation set with fivefold cross-validation.

Results: A total of 500 subjects, including 305 from the RIVUR and 195 from the CUTIE trials, were included in study. Of the subjects 90% were female and mean ± SD age was 21 ± 19 months. A recurrent urinary tract infection developed in 72 patients, of whom 53 also had vesicoureteral reflux (10.6% of the total). The final model included age, sex, race, weight, the systolic blood pressure percentile, dysuria, the urine albumin-to-creatinine ratio, prior antibiotic exposure and current medication. The model predicted recurrent urinary tract infection associated vesicoureteral reflux with an AUC of 0.761 (95% CI 0.714-0.808) in the testing set.

Conclusions: Our predictive model using a novel machine learning algorithm provided promising performance to facilitate individualized treatment of children with an initial urinary tract infection and identify those most likely to benefit from voiding cystourethrogram after the initial urinary tract infection. This would allow for more selective application of this test, increasing the yield while also minimizing overuse.

Keywords: forecasting; machine learning; urinary bladder; urinary tract infection; vesico-ureteral reflux.