Febrile Infants With Urinary Tract Infections at Very Low Risk for Adverse Events and Bacteremia
BACKGROUND: There is limited evidence from which to derive guidelines for the management of febrile infants aged 29 to 60 days with urinary tract infections (UTIs). Most such infants are hospitalized for ≥48 hours. Our objective was to derive clinical prediction models to identify febrile infants with UTIs at very low risk of adverse events and bacteremia in a large sample of patients.
METHODS: This study was a 20-center retrospective review of infants aged 29 to 60 days with temperatures of ≥38°C and culture-proven UTIs. We defined UTI by growth of ≥50 000 colony-forming units (CFU)/mL of a single pathogen or ≥10 000 CFU/mL in association with positive urinalyses. We defined adverse events as death, shock, bacterial meningitis, ICU admission need for ventilator support, or other substantial complications. We performed binary recursive partitioning analyses to derive prediction models.
RESULTS: We analyzed 1895 patients. Adverse events occurred in 51 of 1842 (2.8% [95% confidence interval (CI): 2.1%–3.6%)] and bacteremia in 123 of 1877 (6.5% [95% CI: 5.5%–7.7%]). Patients were at very low risk for adverse events if not clinically ill on emergency department (ED) examination and did not have a high-risk past medical history (prediction model sensitivity: 98.0% [95% CI: 88.2%–99.9%]). Patients were at lower risk for bacteremia if they were not clinically ill on ED examination, did not have a high-risk past medical history, had a peripheral band count of <1250 cells per μL, and had a peripheral absolute neutrophil count of ≥1500 cells per μL (sensitivity 77.2% [95% CI: 68.6%–84.1%]).
CONCLUSION: Brief hospitalization or outpatient management with close follow-up may be considered for infants with UTIs at very low risk of adverse events.
- urinary tract infections
- emergency department
- outpatient therapy
WHAT'S KNOWN ON THIS SUBJECT:
Febrile infants aged 29 to 60 days who have urinary tract infections are typically hospitalized for ≥48 hours. Previous study results have indicated that most patients have benign clinical courses, but small sample sizes have limited the ability to identify those at near-zero risk of adverse events.
WHAT THIS STUDY ADDS:
We derived prediction models in a large sample of patients that identify infants at very low risk for adverse events and at low risk for bacteremia. Shorter hospitalization or outpatient treatment with close follow-up may be feasible for selected patients.
Urinary tract infection (UTI) is the most common serious bacterial illness in febrile infants younger than 60 days of age, occurring in 4% to 10% of these infants.1,–,6 In 1999, an American Academy of Pediatrics guideline recommended outpatient management with oral or parenteral antibiotics for children older than 2 months who have UTIs unless the child seemed “toxic, dehydrated, or unable to take oral intake,” in which case hospitalization was recommended.7 There is little information however, to guide the management of infants aged 29 to 60 days.8,9 Although a recent clinical trial found it feasible to treat febrile infants with UTIs as outpatients with parenteral antibiotics administered in an infusion center,8 most clinicians hospitalize these infants for ≥48 hours of parenteral therapy.10,–,12
There are several reasons why clinicians hospitalize infants aged 29 to 60 days of age who have UTIs, including the unclear risk of long-term renal injury, and concern regarding follow-up. The most important reasons, however, are the concerns of acute adverse events and for missing concomitant bacteremia. Several small studies have assessed the course of febrile infants with UTIs and suggest that otherwise well-appearing infants with or without concomitant bacteremia have benign clinical courses when treated with appropriate antibiotics.9,–,13 Because of small sample sizes, investigators have neither been able to provide a precise estimate of the risk of adverse events or bacteremia nor derive potential multivariable models to identify risk factors for adverse events or bacteremia in young, febrile infants with UTIs.
The aims of this study were to determine the risk of adverse events and bacteremia in a large sample of febrile infants aged 29 to 60 days with UTIs who present to the emergency department (ED), and to derive clinical prediction models for infants at very low risk of adverse events and bacteremia. These data may help clinicians make more informed decisions regarding the need for, or length of hospitalization of these infants.
Study Design and Setting
We performed a retrospective chart review at 20 medical centers that participated in the Pediatric Emergency Medicine Collaborative Research Committee of the American Academy of Pediatrics. The participating centers included 16 tertiary care pediatric EDs and 3 general EDs in the United States and 1 Canadian tertiary care ED. Approval for the study with waiver of informed consent and for data sharing with the coordinating institution and with the centralized data center was granted by the institutional review board at each participating institution.
We performed case ascertainment by querying laboratory databases for all urine cultures with bacterial growth in patients aged 29 to 60 days that were obtained in the ED between January 1995 and May 2006. Lactobacillus, Micrococcus, diptheroids, Bacillus species and Staphylococcus epidermidis were considered contaminants.
Children were classified as having a UTI if urine cultures grew a single pathogen and colony counts met at least 1 of 3 criteria:
≥1000 colony-forming units (CFU)/mL for urine cultures obtained by suprapubic aspiration;
≥50 000 CFU/mL from a catheterized specimen; or
≥10 000 and <50 000 CFU/mL from a catheterized specimen in association with positive urinalysis results.14
We defined positive urinalysis results as those that met any of the following: (a) any organisms visualized on Gram-stain; (b) trace or greater result for leukocyte esterase or nitrite on ED dipstick or laboratory-based urinalysis; or (c) ≥5 white blood cells (WBCs) per high-power field (standard microscopy) or per μL (hemocytometer) on a centrifuged or uncentrifuged urine specimen.15,–,17
We excluded patients for any of the following: (a) transfer from other hospitals with previously obtained laboratory results; (b) urine specimens obtained by techniques other than suprapubic aspiration or transurethral catheterization; (c) urine cultures that grew multiple organisms; or (d) no measured temperature of ≥38.0°C in the ED or at home within 24 hours of ED presentation. We excluded patients without a measured fever because we were interested in patients with a high likelihood of true UTIs, not those with asymptomatic bacteriuria.
Data Collection and Potential Risk Factors
All coinvestigators were trained by the study's principal investigator in person or during conference calls. Each of the coinvestigators reviewed the medical charts for all study patients at their site. Study data were entered onto computerized PDF documents (“teleforms”) (Adobe Acrobat 8 Professional, San Jose, CA) that were programmed to improve data accuracy and completeness. The teleforms were then uploaded electronically or by fax into a central database to avoid secondary transcription errors.
We collected data on patient demographics, past medical history, presenting symptoms in the ED (including upper-respiratory infection symptoms, vomiting, diarrhea, difficulty feeding, and presence of seizures), vital signs at triage, ED physical examination findings (general appearance, dehydration, and respiratory distress), and ED disposition. We obtained data regarding ED and inpatient management and the patient's clinical course, radiologic study results, as well as urine, blood, and cerebrospinal fluid (CSF) laboratory and microbiology results.
We also assessed for the presence of acute concomitant diseases defined as an acute, focal infectious process distinct from the UTI, such as pneumonia, bronchiolitis, cellulitis, osteomyelitis, or septic arthritis. We did not consider acute gastroenteritis or acute otitis media as an acute concomitant disease.
A priori, we created the variable “clinically ill in the ED,” defined as an infant who was judged as ill-appearing, dehydrated, or in respiratory distress or who had an acute concomitant disease diagnosed in the ED. We also created the variable “high risk past medical history,” which included a history of genitourinary abnormalities, previous UTIs, bacteremia, meningitis, previous laboratory evaluation for fever, prematurity (<37 weeks' gestation), or history of a severe systemic disease (complex heart, chronic lung, metabolic, or neurologic diseases). Other past medical history such as minor neonatal complications, jaundice, gastroesophageal reflux or a history of a minor resolved acute illness was not considered to represent a high-risk past medical history (coded “not high risk”).
We had 2 study outcomes: adverse events and bacteremia. Although the occurrence of an adverse event was considered the most clinically important outcome, we also included bacteremia because it is clinically relevant even in the absence of an adverse event.
An adverse event was considered present if any of the following occurred: death, shock, bacterial meningitis, ICU or step-down ICU admission or transfer (for ICU or step-down ICU level of care and monitoring), need for ventilatory support (intubation, continuous, or bilevel positive airway pressure), need for surgical intervention, or other substantial clinical complication. Patients did not meet this definition if the adverse event was solely related to an iatrogenic complication (eg, anaphylaxis because of a medication, nosocomial infection).
We defined shock as (a) “shock” clearly stated in a faculty or fellow physician note, (b) use of vasopressors, or (c) the combination of low blood pressure (<70 mm Hg systolic) or “hypotension” or “sepsis” clearly stated in medical chart (“rule out sepsis” did not qualify as sepsis) and the patient was treated with intravenous fluid boluses of ≥40 mL/kg.
We defined definite bacterial meningitis as the growth of a known pathogen in the CSF. Bacillus species, Propionibacterium acnes, non–Staphylococcus aureus, and Streptococcus viridans were considered contaminants.
For patients with no bacterial growth in the CSF cultures, we defined probable bacterial meningitis as any of the following:
the combination of (a) CSF pleocytosis (≥10 cells per μL) and (b) positive blood culture and (c) treatment consistent with bacterial meningitis (defined as ≥14 days of parenteral antibiotics not clearly administered for other reasons on review of inpatient records); or
the combination of (a) positive CSF Gram-stain or positive latex agglutination tests and (b) treatment consistent with bacterial meningitis; or
the combination of (a) pretreatment with antibiotics before lumbar puncture, (b) CSF pleocytosis (≥10 cells per μL), and (c) treatment consistent with bacterial meningitis.
Bacteremia was defined as the growth of a pathogen in the blood culture. We considered blood cultures that grew Bacillus species, Propionibacterium acnes, or non–S aureus as contaminated.
Data Validity and Reliability
Subjective Physical Examination Variables
To minimize the potential bias associated with abstracting physical examination findings from the medical chart, we used specific, restrictive key words to determine and assign whether the infant was ill-appearing, dehydrated, or in respiratory distress (see Appendix). To determine inter-rater reliability, a second assessor at each institution performed an independent assessment of the documented physical examination for a random sample of 10% of patients as well as all patients who had bacteremia or evidence of adverse events.
Determination of Adverse Events
To achieve agreement on the presence and timing of adverse events, a second investigator at each institution independently reviewed the medical charts of all patients who potentially had adverse events. For these patients, both site investigators assessed (a) the nature of the adverse events and (b) the time points at which the adverse events were identified.
We labeled unavailable data as missing except for particular presenting symptoms (upper-respiratory infection symptoms, vomiting, difficulty feeding, diarrhea, and seizures), for which the absence of a specific description in the chart was interpreted as “not present.”
Patients Discharged Home From the ED
For these patients, investigators completed a detailed chart review of all subsequent visits to the hospital within 1 year of the ED visit at which the UTI was diagnosed. If there were subsequent visits but no documentation of adverse events that occurred during the UTI illness, we categorized the outcome as uncomplicated. If there were no subsequent visits documented within the year, we categorized the outcome as unknown.
We described the risk of adverse events and bacteremia as proportions with 95% confidence intervals (CIs). The unweighted Cohen's κ value was used to determine the interrater reliability for the assessor review of subjective physical examination findings.
Prediction Models for Adverse Events and Bacteremia
We performed 2 separate binary recursive partitioning analyses to identify a group of infants at very low risk for adverse events and a group of infants at very low risk for bacteremia. In these analyses, we included all the following potential predictor variables with biological plausibility for association with the outcome variables: age; past medical history; history of feeding difficulty; vomiting; seizures; clinical appearance in the ED; presence of acute concomitant disease; vital signs at triage; peripheral WBC count; peripheral blood absolute neutrophil count (ANC); peripheral blood band count; peripheral blood immature/total neutrophil ratio; and CSF WBC count. Continuous variables were dichotomized and rounded by analyzing frequency distributions, receiver operating curves, and single-variable recursive partitioning analyses to identify the best predictive and clinically sensible cutoff points for association with each of the 2 outcomes.
Because our aim was to identify which patients were at very low risk of the 2 outcomes, we assigned a high relative cost to misclassification of patients with adverse events or bacteremia. We assigned a relative cost of 100 to 1 for failure to identify a patient with an adverse event versus incorrect classification of a patient without an adverse event, and a cost of 20 to 1 for failure to identify a patient with bacteremia versus incorrect classification of a patient without bacteremia. We used “Gini” splitting rules and pruned the resulting trees to improve simplicity and generalizability. Finally, we used 10-fold cross-validation to develop robust and generalizable prediction models. We report test characteristics for each outcome and 95% CIs, calculated with exact methods. We used Classification and Regression Tree software (CART 6 [Salford Systems, San Diego, CA]) to conduct the recursive partitioning analyses. Finally, we performed bivariate and multivariate regression analyses (using SPSS 16 [SPSS Inc, Chicago, IL]) for variables included in the prediction models to assess the strength of association of these variables with the 2 outcomes.
We based our sample size on the ability to provide precise estimates of the risk of adverse events, the risk of bacteremia, and the sensitivities of the prediction models. In a sample of 1500 infants aged 29 to 60 days with fever and UTIs, we expected the upper ends of the 95% CIs for an assumed 2% estimated risk of adverse events and 5% estimated risk of bacteremia to be 2.8% and 6%, respectively. Given these risk assumptions, we required ∼50 patients with adverse events and 100 patients with bacteremia to result in a 95% CI lower boundary of 94% for a model to predict adverse events that was 100% sensitive, and a lower boundary of 89% for a model to predict bacteremia that was 95% sensitive. Previous literature suggested we were unlikely to derive a model that was 100% sensitive to detect bacteremia.12,13
The information from 1 of the 20 participating institutions was excluded because of systematically missing data. The remaining institutions had access to microbiology databases dating 3 to 10 years. We identified 2477 potentially eligible patients, of whom 1895 were included in the analysis (Fig 1).
The demographics, clinical characteristics, and general laboratory results of our population are summarized in Table 1. Most patients had no significant past medical histories, were febrile for <24 hours, and were well-appearing. Escherichia coli was the predominant organism in the urine. Most patients were hospitalized and treated with parenteral antibiotics, with a median length of hospitalization of 3 days (interquartile range: 2–5 days).
Risk of Adverse Events
Adverse events occurred in 51 of 1842 infants for whom outcome data were available (2.8% [95% CI: 2.1%–3.6%]) (Table 2). Adverse events were diagnosed immediately in the ED in 26 of 51 (51.0%) patients, within 4 hours of ED presentation in 10 of 51 (19.6%), between 4 and 23 hours in 11 of 51 (21.6%), between 24 and 47 hours in 2 of 51 (3.9%), and after ≥48 hours in 2 of 51 (3.9%).
Risk of Bacteremia
Bacteremia was present in 123 of 1877 infants from whom blood cultures were obtained (6.5% [95% CI: 5.5% of 7.7%]) (Table 1). The time from obtaining the blood culture to the time that the blood culture was noted to be positive was available for 91 of 123 patients (74%). The median time to positivity was 16 hours (interquartile range: 13–24 hours); 80 of 91 (88%) were positive within 24 hours. Ten of 123 patients with bacteremia had adverse events (8.1% [95% CI: 4.2%–13.8%]), whereas 41 of 1754 patients without bacteremia had adverse events (2.3% [95% CI: 1.7%–3.1%).
Prediction Models for Adverse Events and Bacteremia
Recursive partitioning analysis identified a group of patients at very low risk of adverse events (Fig 2). Patients belonged to the very low risk group if they were not clinically ill on ED examination (well-appearing, not dehydrated, not in respiratory distress, and no concomitant acute disease) and did not have a high-risk past medical history. The prediction model sensitivity for adverse events was 98.0% (95% CI: 88.2%–99.9%), and the negative predictive value was 99.9% (95% CI: 99.5%–100%). Of 1206 infants in the very low risk group (65.5% of those analyzed), only 1 infant had an adverse event (0.1% [95% CI: 0%–0.4%]): a 46-day-old boy who was well-appearing on physical examination in the ED but whose initial CSF studies were lost. A subsequent lumbar puncture 24 hours after the initiation of intravenous antibiotics revealed a negative Gram-stain, 34 WBCs per μL, 177 red blood cells per μL, 54 mg of glucose per dL, 85 mg of protein per dL, negative bacterial culture, and negative CSF bacterial antigens. In accordance with subspecialist recommendations, the infant was treated for 21 days with intravenous antibiotics for possible pretreated bacterial meningitis. The patient did not have bacteremia and had an otherwise uneventful hospital course.
The recursive partitioning analysis identified patients to be at low risk for bacteremia if they met all of the following criteria: not clinically ill in the ED, no high risk past medical history, a peripheral band cell count of <1250 cells per μL, and a peripheral ANC of ≥1500 cells per μL (Fig 3). The prediction model sensitivity was 77.2% (95% CI: 68.6%–84.1%), and the negative predictive value was 96.8% (95% CI: 95.3%–97.8%). Bacteremia was present in 28 of 862 patients (3.2% [95% CI: 2.2%–4.6%]) who met all the low risk criteria. None of these 28 low-risk infants had an adverse event.
Finally, Tables 3 and 4 demonstrate the unadjusted and adjusted odds ratios for predictor variables in the 2 analyses. The presence of clinical illness in the ED was the variable most strongly associated with adverse events, whereas total peripheral band count ≥1250 cells per μL and peripheral ANC of <1500 cells per μL were most highly associated with bacteremia.
In this large multicenter study, we derived a clinically sensible and parsimonious prediction model that identifies a group of infants aged 29 to 60 days with fever and UTIs who are at very low risk of developing adverse events when treated with antibiotics. The model accurately predicts with very high negative predictive value that infants who are not clinically ill in the ED and have no high-risk past medical history will have benign clinical courses. The model was highly sensitive, misclassifying only 1 patient (0.1%) with an adverse event. This patient's adverse event (bacterial meningitis) could be questioned because the CSF studies obtained in the ED were lost, the subsequent CSF findings were unlikely to represent bacterial illness, and his clinical course was uncomplicated. Although we are aware of no similar studies to derive models to predict adverse events in this population, the overall low risk of adverse events that we found is similar to that in small previous studies of infants with UTIs.8,10,12,13
Our data suggest that infants with UTIs and fever who meet very low risk criteria for adverse events (ie, those who are not clinically ill in the ED and have no high-risk past medical history) could be managed less conservatively than commonly practiced. Possible strategies for managing these infants include hospitalization for short periods of time (eg, 24 hours) or perhaps discharge with close follow-up to a reliable home environment after appropriate antibiotic therapy is initiated and the patient has a period of observation in the ED. In the present study, approximately two-thirds of the infants met the low risk criteria for adverse events. Investigators recently showed that outpatient intravenous antibiotic treatment in an infusion center was feasible and safe in low risk infants with UTIs. That study, however, included only 25 patients aged 29 to 60 days.8 In our series, 176 patients (9.3% of our study population) were discharged home from the ED after receiving parenteral ceftriaxone without any known subsequent complications, including patients later found to have bacteremia. A shorter hospitalization or outpatient therapy would decrease the exposure of these infants to the potential iatrogenic complications associated with hospitalization.21
We recognize that clinicians may be unwilling to discharge home from the ED febrile infants with UTIs who potentially have bacteremia, even if they are likely to have otherwise uncomplicated clinical courses. In our series, 6.5% of patients had bacteremia, whereas previous study estimates range from 0% to 21%.9,10,12,22,–,24 Similar to previous literature, most patients in our study who had positive blood cultures were identified within the first 24 hours.25 We attempted but were unsuccessful in deriving a very low risk model to identify infants with bacteremia. This reflects the limitations in predicting bacteremia with standard clinical and laboratory data.12,13 The addition of a peripheral band count of <1250 cells per μL and peripheral ANC of ≥1500 cells per μL to those predictors that accurately identify infants at very low risk for adverse events did decrease the likelihood of bacteremia to 3.2%. Previous research has noted an association between high peripheral band counts or low peripheral ANC with bacteremia caused by Gram-negative organisms.12,23,26,27
Our study had several limitations typical of medical chart reviews. These included potential biases in data abstraction, particularly of subjective clinical findings. We attempted to minimize these biases by creating a detailed manual of operations that included specific key words to interpret subjective findings, by conducting interrater reliability analyses of subjective variables, and by using more objective criteria to define adverse events. These efforts would not account for clinician documentation biased by previous knowledge of laboratory results or clinical course. However, clinicians in the ED would be unlikely to know of future clinical deterioration before initial documentation in the medical chart.
We were also challenged to define bacterial meningitis for patients with negative CSF cultures. We used conservative definitions for probable bacterial meningitis to avoid missing patients with this important outcome. It should also be recognized that we queried microbiology databases rather than attempting to identify patients who presented to the ED with positive urinalyses, which was not feasible. Therefore, our results are applicable to those patients for whom urine culture results are known and cannot necessarily be extrapolated to those for whom only preliminary screening tests for UTI are known.
We derived a highly accurate prediction model that identifies a group of febrile infants aged 29 to 60 days with UTIs at very low risk for adverse events. We attempted but were unsuccessful in deriving a very low risk model to identify infants who had bacteremia. Initiating antimicrobial therapy and brief hospitalization (eg, 24 hours), within which time frame most bacteremia will be identified, seems appropriate management for this group of infants. Outpatient management with long-acting intramuscular antibiotics and close follow-up could also be considered after a period of observation. Future research should attempt to validate the prediction model for adverse events and continue to assess the safety and feasibility of alternative management strategies for these young febrile infants with UTIs.
The authors would like to thank John Kanegaye, MD, Ma Long and Achilles Kalnoky, MD, for their participation and efforts in this study.
- Accepted August 16, 2010.
- Address correspondence to David Schnadower, MD, MPH, Division of Pediatric Emergency Medicine, Washington University School of Medicine, 660 S Euclid Ave, Campus Box 8116, St. Louis, MO 63130. E-mail:
This work was presented at the annual meetings of the American Academy of Pediatrics National Conference and Exhibition; October 10, 2008; Boston, MA; and October 16, 2009; Washington, DC.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
- UTI =
- urinary tract infection •
- ED =
- emergency department •
- CFU =
- colony-forming unit(s) •
- WBC =
- white blood cell •
- CSF =
- cerebrospinal fluid •
- CI =
- confidence interval •
- ANC =
- absolute neutrophil count
- Baker MD,
- Bell LM,
- Avner JR
- Bonadio WA,
- Smith DS,
- Sabnis S
- Shaw KN,
- Gorelick M,
- McGowan KL,
- Yakscoe NM,
- Schwartz JS
- Zorc JJ,
- Kiddoo DA,
- Shaw KN
- 7.↵American Academy of Pediatrics, Committee on Quality Improvement and Subcommittee on Urinary Tract Infection. Practice parameter: the diagnosis, treatment, and evaluation of the initial urinary tract infection in febrile infants and young children. Pediatrics. 1999;103(4 pt 1):843–852
- Doré-Bergeron MJ,
- Gauthier M,
- Chevalier I,
- McManus B,
- Tapiero B,
- Lebrun S
- Hoberman A,
- Wald ER,
- Hickey RW,
- et al
- Bachur R
- Zorc JJ,
- Levine DA,
- Platt SL,
- et al
- Gorelick MH,
- Shaw KN
- Shaw KN,
- McGowan KL,
- Gorelick MH,
- Schwartz JS
- Herr SM,
- Wald ER,
- Pitetti RD,
- Choi SS
- Nigrovic LE,
- Malley R,
- Macias CG,
- et al
- Crain EF,
- Gershel JC
- Ginsburg CM,
- McCracken GH Jr.
- Kaplan RL,
- Harper MB,
- Baskin MN,
- Macone AB,
- Mandl KD
- Copyright © 2010 by the American Academy of Pediatrics