Clinical characteristics and prediction analysis of pediatric urinary tract infections caused by gram-positive bacteria

Gram-positive (GP) pathogens are less accounted for in pediatric urinary tract infection (UTI), and their clinical impact is underrecognized. This study aimed to identify predictors of GP uropathogens in pediatric UTI. In this 14-year retrospective cohort of pediatric patients with UTI, we classified first-time UTIs cases into those caused by GP or Gram-negative (GN) bacteria. We constructed a multivariable logistic regression model to predict GP UTI. We evaluated model performance through calibration and discrimination plots. We developed a nomogram to predict GP UTI that is clinically feasible. Of 3783 children with first-time UTI, 166 (4.4%) were infected by GP and 3617 (95.6%) by GN bacteria. Among children with GP UTI, the most common uropathogens were vancomycin-resistant Enterococcus faecalis (VRE) (27.1%), Staphylococcus saprophyticus (26.5%), and coagulase-negative Staphylococci (12.7%). Eight independent risk factors were associated with GP UTI: Age ≥ 24 months (odds ratio [OR]: 3.21), no prior antibiotic use (OR: 3.13), serum white blood cell (WBC) count < 14.4 × 103/μL (OR: 2.19), high sensitivity C-reactive protein (hsCRP) < 3.4 mg/dL (OR: 2.18), hemoglobin ≥ 11.3 g/dL (OR: 1.90), negative urine leukocyte esterase (OR: 3.19), negative urine nitrite (OR: 4.13), and urine WBC < 420/μL (OR: 2.37). The model exhibited good discrimination (C-statistic 0.879; 95% CI 0.845–0.913) and calibration performance. VR E. faecalis, the leading GP uropathogen causing pediatric UTI, requires early detection for infection control. Our model for predicting GP UTI can help clinicians detect GP uropathogens and administer antibiotic regimen early.

Study population. From 2003 through 2016, we identified 28,874 paired urinalysis (UA) and urine culture (UC) samples obtained from pediatric patients (age ≤ 18 years) at CMUH, and 26,066 UA-UC pairs were obtained for the same visit from the same patient 13 (Fig. 1). One UA was paired with one UC, which was ordered within 7 days after and closest to the time of UA. In Taiwan, the physician's order of UA and UC exam is supervised by the National Health Insurance (NHI) Administration and can only be reimbursed if the UA and UC exam follow symptoms of UTI (e.g., fever, dysuria, urgency, frequency, incontinence, abdominal, back, or flank pain, nausea, vomiting, poor feeding, irritability, jaundice, or weight loss) 14 .
We classified paired UA-UC cases into positive for UTI (N = 6348) and negative for UTI (N = 19,718) based on sampling source-specific cutoffs of colony forming units (CFUs). UTI is defined as urine culture containing ≥ 10 5 CFU/mL in a midstream urine specimen, ≥ 10 4 CFU/mL in a catheter urine specimen, or ≥ 10 3 CFU/mL in a percutaneous nephrostomy (PCN) or suprapubic urine specimen, based on the EAU/ESPU guidelines 15 . The unclassified urine cultures, including those with no growth or with contamination (i.e., presence of more than three organisms) were classified as negative for UTI. Moreover, children included as UTI positive all received antibiotics and the antibiotic treatment for inpatients were routinely approved by pediatric infectious disease specialist and reviewed by the NHI Administration for reimbursement.
To compare the characteristics between patients with GP and GN pathogens, our study population was formed from 6348 UTI positive cases and divided into UTI caused by GP or GN pathogens. In addition, we excluded patients if their urine WBC count was less than 25/μL, if their urine culture grew both GP and GN pathogens, if they had catastrophic illness as defined by the Ministry of Health and Welfare, Taiwan 16 , and if the time interval between UA and UC pairs was ≥ 3 days. Our study population, consisting of 3783 children with UTIs-3617 with GN UTI and 166 with GP UTI-were used in all subsequent analyses.
Covariates. The details of urinalysis and urine culture methods are described in "Supplemental Text". Prematurity was defined using the ICD-9 codes of 765.xx that were presented in the EHR prior to the UA order. Hydronephrosis (ICD-9 codes 591.xx), vesicoureteral reflux (ICD-9 code 593.7x), and hypospadias (ICD-9 code 752.6x) were defined using ICD-9 codes presented in the EHR within 1 year prior to the UA order. Results from imaging studies, such as kidney sonography, voiding cystoureterography, and Tc 99 m-dimercaptosuccinic acid renal scintigraphy, performed within 1 year prior to the UA order, were evaluated. History of Foley catheteriza-  www.nature.com/scientificreports/ tion was defined as placement of Foley catheter within 3 months prior to the UA order. History of bacteremia was defined as having positive blood culture within 3 months prior to the UA order. Serum biochemistry profiles for WBC, Hb, platelets, and high sensitivity C-reactive protein (hsCRP), which were measured within 7 days prior to the UA order, were analyzed. Continuous variables were dichotomized using the median value as the cutoff.
Outcome variables. Length of stay (LOS) was defined as the duration between admission and discharge for patients who were admitted to the CMUH. Recurrent UTI was defined as having ≥ 2 UTIs within 6 months or having ≥ 3 UTIs within 1 year following the index UA. Bacteremia was defined as having positive blood culture within 90 days following the index UA order.
Statistical analysis. Descriptive statistics are presented as mean (standard deviation) and median (interquartile range) for continuous variables and as frequency and proportion (%) for categorical variables. We compared characteristics between patients with UTI caused by GP and GN bacteria by using the Wilcoxon rank-sum test or the chi-square test. We established a multivariable logistic regression model using statistically significant or clinically relevant variables to predict GP UTI. We examined the discrimination and calibration performance of the model by using the c-statistic, the receiver-operating curve, and the calibration plot 17 . To maximize clinical utilization, we calculated the risk points on the basis of risk estimate and developed a nomogram by using R with the rms package 18 . Decision curve analysis was used to evaluate the clinical net benefit of our prediction model 19 . All analyses were performed using SAS Version 9.4 (Cary, NC, USA. https:// www. sas. com) or R Version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria. https:// www.r-proje ct. org). The significance level was set at 0.05, and all tests were two-tailed.

Discussion
In this 14-year hospital-based cohort study, we found that the distribution of GP UTI was stable over the study period, with GP bacteria contributing to approximately 4.4% of all pediatric UTI events. The top three pathogens for GN and GP uropathogens were E. coli, P. mirabilis, and K. pneumoniae, and vancomycin-resistant E. faecalis (VRE), S. saprophyticus, and coagulase-negative Staphylococci, respectively. VRE was the causative GP uropathogen in children younger than 2 and between 2 and 11 years of age; however, S. saprophyticus was predominant in children older than 12 years of age. Our prediction model for GP UTI in children has both good discrimination and calibration and the nomogram can make clinicians aware of the potential GP uropathogens.
The leading GP uropathogens and their distribution across age groups found in our study were consistent with that in the literature. Enterococcus spp. is the most common GP uropathogen in the pediatric outpatient population in the US 20 23 . VRE is a rapidly emerging multidrug-resistant pathogen causing infection in adults since its discovery in 1986 24 . A nationwide study of hospitalized children in the United States documented that VRE infection increased from 53 per million in 1997 to 120 per million in 2012 25 . As VRE poses a critical threat to hospital infection control, our findings facilitate risk management of pediatric UTI and inform infection control policy in children, especially those younger than 2 years of age, with GP UTI.
Marrow responses may help differentiate GP UTI from GN UTI. In our study, children with GP UTI had a lower WBC and platelet count and lower hsCRP level. This finding indicates that GP UTI has a lower inflammatory response, which is consistent with previous studies suggesting a more profoundly elevated WBC count and erythrocyte sedimentation rate in E. coli UTI compared with non-E. coli UTI in children 26,27 . The absence of urine nitrate or leukocyte esterase and the presence of pyuria indicated higher odds of GP UTI in our study and in two other studies 5,28 . Decreased levels of inflammatory biomarkers in the serum and urine in GP UTI may be Table 1. Demographic and clinical characteristics of 3783 pediatric patients with urinary tract infection, 2003-2016 at China Medical University Hospital (CMUH), Taiwan. Urinary tract infection was defined as growth of bacteria at a concentration of at least 10 5 CFU/mL in a midstream urine sample, 10 4 CFU/ mL in a catheter urine sample, or 10 3 CFU/mL in a PCN urine sample or suprapubic urine sample. Only UTIs caused by a single strain of a uropathogen were included. CFU colony-forming unit, DMSA Tc 99 m-dimercaptosuccinic acid, ER emergency room, hsCRP high-sensitivity C-reactive protein, IQR interquartile range, PCN percutaneous nephrostomy, RBC red blood cell, SD standard deviation, UA urinalysis, UC urine culture, VCUG voiding cystoureterography, VUR vesicoureteral reflux, WBC white blood cell. a Categorical variables are presented as frequency (%) and continuous variables are presented as median (Q1, Q3), if not otherwise specified. b To compare the difference between GPC and GNB, the p value was calculated using the chi-square test or Wilcoxon rank-sum test. c Occurred any time prior to or on the day of the index UA order. d Occurred within 1 year prior to or on the day of the index UA order. e Occurred within 3 months prior to or on the day of the index UA order. f The latest measure within 7 days prior to or on the day of the index UA order. g Length of hospitalization was calculated for 939 patients (903 GN; 36 GP) who were hospitalized. h Recurrent UTI was defined as having ≥ 2 UTIs within 6 months or having ≥ 3 UTIs within 1 year. i Unilateral and bilateral ureteroneocystostomy that was performed anytime following the UA order. j Bacteremia that occurred within 90 days following the UA order.  28 . The rate of ureteroneocystostomy, an operation to correct VUR, is higher in children with GP UTI. Pediatric non-E. coli UTI is associated with anatomical abnormalities, specifically VUR 26,27,30 . Enterococcus, S. aureus, and coagulase-negative Staphylococci are associated with VUR possibly because the urinary tract abnormalities allow low virulence GP bacteria to attach 9,10,30 . The common approaches to VUR include vigilant observation, antibiotic prophylaxis, and surgical correction, which is indicated in children with persistent VUR, recurrent UTI under antimicrobial prophylaxis, or high-grade reflux 15,31 .
Delayed treatment for febrile UTIs is significantly associated with permanent renal scarring 3 . Cefazolin, the first-line empirical therapy for UTIs, frequently fails to cover UTIs caused by GP bacteria and thus can increase the risk of renal scarring, especially among children with GP pyelonephritis 6,9,27,29,30 . Therefore, our prediction model for GP UTI can guide physicians to initiate appropriate antibiotic treatment early.
The strengths of our study include objective guideline-approved UTI definition using colony count and culture source, large sample size of 3783 pediatric UTIs, robust multivariable analysis to establish a prediction model for GP UTI, and addition of serum biomarkers to the prediction model for GP UTI. Our study also had a few limitations. First, our prediction model for GP UTI cannot be applied to UTIs with mixed GP and GN bacteria because our model was developed using GP-and GN-only UTIs. Second, some of the unmeasured features, such as history of UTI, prior antibiotic treatment, or other clinical histories, that were documented in other www.nature.com/scientificreports/ institutions, could have affected the likelihood of GP UTI. Third, our prediction model, developed in a tertiary medical center in central Taiwan, may not be applicable to other healthcare facilities. However, as one of the largest tertiary medical centers in Taiwan, our patient population should be representative. This is the first study to establish a prediction model for GP UTI in a pediatric population. Age older than 2 years, no prior antibiotic use, low blood and urine WBC count, low hsCRP level, high hemoglobin level, and absence of urine nitrite and leukocyte esterase are significant predictors of pediatric UTI caused by GP bacteria. Our prediction model for GP UTI in children could help clinicians quantify the probability of infection by GP uropathogens and enable them to choose an adequate antibiotic regimen early. Large prospective studies in the future should validate our findings.