Abstract
OBJECTIVE: To validate externally the UTICalc, a popular clinical decision support tool used to determine the risk of urinary tract infections (UTIs) in febrile children, and compare its performance with and without the inclusion of race and at differing risk thresholds. METHODS: We performed a retrospective, singlecenter case-control study of febrile children (2-24 months) in an emergency department. Cases with culture-confirmed UTI were matched 1:1 to controls. We compared the performance of the original model which included race (version 1.0) to a revised model which did not consider race (version 3.0). We evaluated model performance at risk thresholds between 2% and 5%. RESULTS: We included 185 cases and 197 controls (median age 8.4 months; IQR, 4.4-13.0 months; 60.5% girls). When using UTICalc version 1.0, the model area under the receiver operator characteristic curve (AUROC) was 73.4% (95% CI 68.4%-78.5%), which was similar to the version 3.0 model (73.8%; 95% CI 68.7%-78.8%). When using a 2% risk threshold, the version 3.0 model demonstrated a sensitivity of 96.7% and a specificity of 25.0%, with declines in sensitivity and gains in specificity at higher risk thresholds. Version 1.0 of the UTICalc had 12 false negatives, of whom 10 were Black (83%); whereas version 3.0 had 6 false negatives, of whom 2 were Black (33%). CONCLUSIONS: Versions of the UTICalc with and without race had similar performance to each other with a slight decline from the original derivation sample. The removal of race did not adversely affect the accuracy of the UTICalc.