Parenteral Antibiotic Therapy Duration in Young Infants With Bacteremic Urinary Tract Infections

Study Design

This study was nested within a multicenter retrospective cohort study^12,13 of young infants evaluated for invasive bacterial infections in the emergency departments (EDs) of 11 geographically diverse children’s hospitals (Supplemental Table 3). The study was approved by each site’s institutional review board with a waiver of informed consent and permission for data sharing.

Study Population

Infants ≤60 days old who presented to the EDs of the 11 participating hospitals between July 1, 2011, and June 30, 2016, were included in the parent study if they had a bacterial pathogen isolated in blood or cerebrospinal fluid (CSF) culture. Pathogenic bacteria were determined a priori via literature review and expert consensus (Supplemental Table 4).^12,13 For this substudy, we included infants from the parent study who were diagnosed with a UTI and had detection of the same bacterial pathogen in both blood and urine cultures (Fig 1). UTI was defined as growth of ≥50 000 colony-forming units (CFUs)/mL of a pathogen in urine culture or growth of ≥10 000 CFUs/mL of a pathogen in urine culture plus an abnormal urinalysis (defined as presence of leukocyte esterase, nitrites, or >5 white blood cells per high-power field on urine microscopy).^4,14,15 We excluded infants who had a bag or unknown collection method for urine culture if pathogen growth was <100 000 CFUs/mL or if multiple organisms were isolated because these organisms could represent contaminants.^16,17 Infants who had concomitant bacterial meningitis, infants with CSF pleocytosis who received antibiotic pretreatment, and infants with examination and laboratory findings consistent with focal infections, such as septic arthritis, were also excluded from our cohort because these infants would likely be prescribed prolonged courses of parenteral antibiotics for treatment of other invasive infections. Finally, we excluded infants who died and those who were transferred out from the participating institution before completion of parenteral antibiotic therapy because we could not report total duration of IV/IM antibiotic therapy in these patients.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Flowchart of the study population. ^a UTI definition: positive urine culture result with ≥50 000 CFUs/mL or ≥10 000 CFUs/mL with an abnormal urinalysis (leukocyte esterase-positive or nitrite-positive or >5 WBCs). ^b Patients who had a bag or unknown method of urine culture collection were excluded if the culture grew <100 000 CFUs/mL or if >1 organism was isolated. WBC, white blood cell.

Data Sources

Each site’s electronic medical record system or microbiology laboratory was queried to identify all infants who had a positive blood or CSF culture result collected in the ED during the study period. A medical record review was then conducted to confirm eligibility. Information on patient demographics, clinical variables (eg, presence of fever and physical examination findings), laboratory data, parenteral antibiotic duration, adverse events, and 30-day ED revisit or readmission were also obtained via detailed medical record review.

Exposure

The exposure of interest was duration of parenteral antibiotic therapy. Patients were divided into short-course and long-course exposure groups. Given that we found a somewhat bimodal distribution of parenteral antibiotic duration with a median duration of 7 days (Supplemental Fig 2), we defined short course as receipt of parenteral antibiotics for ≤7 days and long course as receipt of parenteral antibiotics for >7 days.

Outcomes

The primary outcome was recurrence of UTI that resulted in an ED revisit or hospitalization within 30 days of discharge from the index hospital admission. For infants who had recurrent UTI, we conducted additional medical record review to obtain information on timing of recurrence, causative organism at recurrence, presence of bacteremia or meningitis at recurrence, diagnosis of vesicoureteral reflux (VUR), and use of prophylactic antibiotics before recurrence. Secondary outcomes included hospital length of stay (LOS), 30-day all-cause reutilization (defined as hospital readmission or ED revisit within 30 days of index discharge), and adverse events within 30 days of discharge (ie, readmission to an ICU, need for mechanical ventilation or vasopressor use, or neurologic sequelae).

Covariates

Covariates of interest included patient age, sex, prematurity (defined as gestational age <37 weeks), complex chronic condition,¹⁸ known genitourinary anomaly, presence of fever (ie, a reported temperature ≥38.0°C (100.4°F) at home, in an outpatient clinic, or documented in the ED and obtained via any method), ill appearance, ICU stay during the index admission, causative pathogen, and prolonged bacteremia. Ill appearance was defined as any of the following physical examination terms documented on arrival to the ED: “ill appearing,” “toxic,” “limp,” “unresponsive,” “gray,” “cyanotic,” “apnea,” “weak cry,” “poorly perfused,” “grunting,” “listless,” “lethargic,” or “irritable.”¹⁹ If none of these terms were documented, the infant was classified as not ill appearing. In cases of contradicting documentation of ill appearance between attending physician and trainee, the attending physician’s documentation was used. Prolonged bacteremia was defined as isolation of the initial pathogen on a repeat blood culture drawn at least 24 hours after the initial culture.²⁰

Statistical Analysis

Categorical variables were described by using frequencies and percentages. To adjust for potential confounding by indication for duration of parenteral antibiotic therapy, we generated a propensity score for each patient to estimate the conditional probability of receiving long-course parenteral antibiotic therapy.^21,22 Propensity scores were calculated by using a multivariable logistic regression model that consisted of baseline demographic characteristics (ie, age and sex), past medical history (ie, prematurity, presence of chronic complex condition, and known genitourinary anomaly), and clinical characteristics at presentation (ie, fever, ill appearance, ICU stay, prolonged bacteremia, and causative organism).

We then conducted inverse probability weighting (IPW), aiming to achieve balance in the distribution of covariates between exposure groups by multiplying each study observation by the inverse of its respective propensity score to achieve a statistically balanced pseudo-study population.²¹ Distributional balance of covariates between short-course and long-course antibiotic groups after the weighting procedure was examined by using standardized mean differences, with the magnitude of ≤0.1 indicating acceptable balance (Supplemental Table 5).²¹ One variable, prolonged bacteremia, remained unbalanced; however, it was neither significantly associated with the outcomes of recurrent UTI and reutilization, as tested by Fisher’s exact test, nor with LOS, as tested by the Wilcoxon rank test. To incorporate the effect of study site, we attempted 2 additional models to generate propensity scores: 1 in which study site was used as a fixed-effect predictor in the above logistic regression and a second in which a random intercept was added for each study site. However, neither model yielded adequate balance between covariates. Thus, the multivariable logistic regression model without study site (Supplemental Table 5) was used for the analysis.

To evaluate the association of antibiotic duration with our outcomes of interest, we employed marginal structural models and applied them to our weighted population.²³ We used the logit link function to estimate the probability of recurrent UTI and reutilization in each group and summarized our findings using both relative risk and risk difference. We used the identity link to estimate the adjusted difference in the mean LOS between short-course and long-course treatment groups. Natural log transformation was applied to LOS before its modeling in the marginal structural models. Statistical significance was established at the 2-sided α level of .05. Data were analyzed by using R version 3.4.3 statistical software²⁴ and by using the following statistical packages: lme4,²⁵ tableone,²⁶ ipw,²⁷ sandwich,²⁸ and survey.²⁹