Model for predicting high-grade vesicoureteral reflux in young children presenting with febrile urinary tract infection

NUM-score: A clinical-analytical model for personalised imaging after urinary tract infections

AIM

To identify predictive variables and construct a predictive model along with a decision algorithm to identify nephrourological malformations (NUM) in children with febrile urinary tract infections (fUTI), enhancing the efficiency of imaging diagnostics.

METHODS

We performed a retrospective study of patients aged <16 years with fUTI at the Emergency Department with subsequent microbiological confirmation between 2014 and 2020. The follow-up period was at least 2 years. Patients were categorised into two groups: 'NUM' with previously known nephrourological anomalies or those diagnosed during the follow-up and 'Non-NUM' group.

RESULTS

Out of 836 eligible patients, 26.8% had underlying NUMs. The study identified six key risk factors: recurrent UTIs, non-Escherichia coli infection, moderate acute kidney injury, procalcitonin levels >2 μg/L, age <3 months at the first UTI and fUTIs beyond 24 months. These risk factors were used to develop a predictive model with an 80.7% accuracy rate and elaborate a NUM-score classifying patients into low, moderate and high-risk groups, with a 10%, 35% and 93% prevalence of NUM. We propose an algorithm for approaching imaging tests following a fUTI.

CONCLUSION

Our predictive score may help physicians decide about imaging tests. However, prospective validation of the model will be necessary before its application in daily clinical practice.

Vesicoureteral Reflux: Current Care Trends and Future Possibilities

Although investigations over the past 2 decades have improved our understanding of the natural history of vesicoureteral reflux (VUR) and helped identify those at higher risk of both VUR itself as well as its potential severe sequelae, debate exists regarding key aspects of care, including when to perform diagnostic imaging and which patients benefit from continuous antibiotic prophylaxis. Artificial intelligence and machine learning have the potential to distill large volumes of granular data into practical tools that clinicians can use to guide diagnosis and management decisions. Surgical treatment, when indicated, remains highly effective and is associated with low morbidity.