Shortened IV Antibiotic Course for Uncomplicated, Late-Onset Group B Streptococcal Bacteremia

Study Design and Setting

We identified a retrospective cohort of infants with uncomplicated, late-onset GBS bacteremia using the Pediatric Health Information System (PHIS) database. The PHIS database provides deidentified information that details hospital and patient demographic data, administrative data, such as discharge diagnosis and procedure coding, and detailed billing information, including laboratory, imaging, pharmacy, and supply charge data, for patients who are cared for at 49 US stand-alone children’s hospitals. Patients are assigned a single identifier, which allows for longitudinal linkage of all visits within a single PHIS hospital. Data for the current study were abstracted from emergency, inpatient, or observation visits. The PHIS is operated by the Children’s Hospital Association (Lenexa, KS).

Participants

Patients ≤4 months of age who were discharged from a PHIS hospital between January 2000 and September 2015 with an International Classification of Diseases, Ninth Revision (ICD-9) discharge diagnosis code for GBS disease and bacteremia were eligible for inclusion (see Supplemental Table 3 for ICD-9 code definitions). We did not include encounters after the implementation of International Classification of Diseases, 10th Revision coding in October 2015 because the inclusion and exposure codes that we used were validated only for ICD-9 codes. To limit the cohort to infants with uncomplicated GBS disease for whom providers would be most likely to consider shorter courses of IV antibiotics, infants were excluded if any of the following were present during their index admission: (1) a diagnosis of invasive bacterial or viral infections that might preclude transition to oral therapy (meningitis, osteomyelitis, septic joint, concomitant non-GBS bacterial infection, or herpes simplex virus infection), (2) a history of prematurity (gestational age <29 weeks or birth weight <1500 g), or (3) a complicated or severe hospital course (index hospitalization >14 days or receipt of care in an ICU). Infants <7 days old at the time of admission (early-onset GBS disease) were excluded. Patients who were transferred out of the PHIS hospital were excluded to mitigate outcome ascertainment bias.

Exposure Classification

The exposure of interest was receipt of a shortened course of IV antibiotic therapy. We defined a shortened course as IV antibiotic treatment ≤8 days, which is 2 days shorter than the guideline-recommended minimum duration.⁴ We subtracted 2 days from the guideline minimum to more conservatively classify a shortened course exposure because it is possible that patients could have received an additional day or 2 of parenteral antibiotics before or after their hospitalization (ie, with intramuscular ceftriaxone). The PHIS database details daily medications during hospitalization, but it does not provide information about intended therapy after hospitalization. Therefore, patients with a procedure code (38.93) or a hospital charge for a PICC were assumed to have completed a prolonged course of IV antibiotics regardless of the measured duration while hospitalized.

Accuracy of Inclusion and Exposure Codes

Using a manual medical record review of infants with blood cultures that were positive for GBS at Primary Children’s Hospital (a PHIS center) as the gold standard, we estimated the sensitivity and positive predictive values of the study’s ICD-9 inclusion codes to detect true GBS bacteremia and determined how frequently these patients received antibiotic therapy before hospital admission.

We validated our definition of PICC exposure using a PHIS data set from a previously published study of children who were hospitalized for complicated pneumonia and whose medical records were manually reviewed.¹⁵ We calculated positive and negative predictive values for PICC status in the complicated pneumonia data set according to (1) presence of a PICC by PHIS charge or procedure code (our study definition) or (2) presence of a PICC on chart review.

Outcomes

The primary study outcome was a recurrence of GBS disease, defined as a hospital revisit with a discharge diagnosis code for GBS disease and a discharge diagnosis code for bacteremia, meningitis, or osteomyelitis in a patient’s first year of life. Treatment failure, defined as a recurrence of GBS disease within 14 days of hospital discharge, was a secondary outcome. To account for the risk of misclassification of GBS disease recurrence, 1 author (E.R.C) manually reviewed all of the discharge diagnosis codes for patients who revisited the hospital in the first year of life but were not classified as suffering from a recurrence of GBS disease. Patients with ICD-9 codes for bacteremia, osteomyelitis, or meningitis but no accompanying code for GBS disease and patients with a code for GBS disease but no accompanying code for bacteremia, osteomyelitis, or meningitis were classified as possibly suffering from a recurrence. A PICC complication was defined by using a previously established set of ICD-9 codes that specified infection or mechanical complication of a vascular device (see Supplemental Table 3).¹⁷ In contrast to the original publication, we did not include ICD-9 codes for fever because we did not consider this diagnosis alone as sufficient to define a PICC complication.

To account for the possibility of readmissions for true PICC complications being misclassified as readmissions for GBS disease (falsely elevating the measured rate of recurrence among the prolonged IV therapy group), we divided the prolonged IV therapy group into 2 subgroups for further description. One subgroup received a PICC and was treated with ≤8 days of IV antibiotics as inpatients (these patients were likely discharged with their PICCs). The second subgroup did not receive a PICC and was treated with >8 days of IV antibiotics as inpatients (these patients likely completed their antibiotic treatment in the hospital).

Covariates

Covariates included sex, age (in days), race and/or ethnicity (non-Hispanic white, non-Hispanic African American, Hispanic, or other), primary insurance payer (government, commercial or self-pay, or unknown), history of a birth weight of 1500 to 2499 g or a gestational age of 29 to 36 weeks, history of a complex chronic condition, presence of a concomitant urinary tract infection, hospital case mix index, and the year in which a patient was admitted to the hospital. Complex chronic conditions are defined by ICD-9 codes and derived from hospitalization patterns of children with costly illnesses and congenital defects.¹⁸ The PHIS case mix index is determined as the average cost weight after each discharge from a PHIS hospital is assigned an All-Patient–Refined Diagnosis-Related Group and severity level.

Statistical Analysis

To account for nonrandom treatment selection, we estimated the probability of each patient receiving a shortened IV course. Propensity scores were computed with a multivariable logistic regression, with the predicted probability of a shortened course as the outcome and the above covariates as predictor variables. Patients were then weighted by the inverse of their probability to receive the treatment they actually received. Inverse probability weighting maximizes patient inclusion in the model and can create more balanced comparison cohorts. Finally, the association between exposure and patient outcomes was tested by using logistic regression, with the inverse probability weights assigned to each patient. The association between receipt of a shortened IV antibiotic course and time to a GBS recurrence was measured with an inverse probability–weighted γ regression model, which is robust to nonnormally distributed outcomes. Analyses were performed by using Stata version 14 (Stata Corp, College Station, TX) and R statistical software.¹⁹