Comparison of Antibiotic Prescribing for Pediatric Community-Acquired Pneumonia in Children’s and Non-Children’s Hospitals

Abstract

This analysis uses hospital billing databases of inpatient discharges to compare antibiotic prescribing for community-acquired pneumonia by hospital type.

Pneumonia is the most common indication for antibiotic use in hospitalized children.¹ National guidelines for community-acquired pneumonia (CAP) recommend penicillin, amoxicillin, and ampicillin as first-line agents for children hospitalized with CAP.² Although use of these agents for CAP is increasing in children’s hospitals,³ antibiotic prescribing trends for children admitted to non-children’s hospitals, where 70% of hospitalized children receive care,⁴ are unknown. We compared antibiotic prescribing for CAP between hospital types before and after guideline publication in 2011.

Methods

We performed a retrospective analysis using 2 hospital billing databases of inpatient discharges from across the United States: Pediatric Health Information System (PHIS, children’s hospitals only) and Premier Perspectives (all hospital types). We included children (aged 1-17 years) admitted for CAP and discharged between January 1, 2009, and September 30, 2015, with a primary diagnosis of pneumonia (previously validated algorithm⁵), receipt of a systemic antibiotic potentially prescribed for CAP (penicillins with or without β-lactamase inhibitors, cephalosporins, carbapenems, macrolides, doxycycline, fluoroquinolones, vancomycin, linezolid, clindamycin, or trimethoprim-sulfamethoxazole), and an overnight stay. We excluded children with complex chronic conditions,⁵ severe and/or complicated CAP (hospitalized >7 days; effusion, empyema, or lung abscess; 2 consecutive days of mechanical ventilatory assistance or vasopressor use; or extracorporeal membrane oxygenation), and infection, colonization, or history of methicillin-resistant Staphylococcus aureus. Premier Perspectives contains deidentified data, and PHIS data are not readily identifiable. The Children's Hospital of Philadelphia Institutional Review Board has determined that analysis using PHIS is not human subjects research.

To distinguish children’s hospitals from non-children’s hospitals in Premier Perspectives, we applied a definition informed by PHIS using the observed minimum proportion of all discharges (≥1 year of age) that were for children (PHIS minimum during study period, 49%; Premier Perspectives hospitals with ≥49% were classified as children’s hospitals). Two Premier Perspectives children’s hospitals were identified as likely duplicates of PHIS hospitals based on census region, annual discharges, and annual CAP cases and were excluded. Using a longitudinal piecewise logistic model with a knot at October 2011 (guideline publication), we modeled the linear trajectory of guideline-concordant prescribing (defined as receipt of any penicillin, amoxicillin, or ampicillin during admission) before and after publication, stratified by hospital type (children’s vs non-children’s). Potential confounders (age, sex, and insurance payer) did not meaningfully affect model estimates and were excluded from the final model. We adjusted SEs for clustering by hospital. Probabilities of guideline-concordant prescribing with 95% CIs were calculated (CI by Δ method) from the fitted model. Two-sided P < .05 from Wald tests was considered statistically significant.

Results

We identified 120 238 children (62 923 boys [52.3%]; median age, 3 years [interquartile range, 1-6 years]) with CAP discharged from 51 children’s hospitals (46 PHIS hospitals, 5 Premier Perspectives hospitals; 65 209 children [54.2%]) and 471 non-children’s hospitals (55 029 children [45.8%]). At both hospital types, the trajectory of guideline-concordant prescribing changed after guideline release (Figure). In children’s hospitals, the modeled probability of guideline-concordant prescribing increased from 0.25 (95% CI, 0.15-0.34) immediately before guideline release to 0.61 (95% CI, 0.56-0.66) at study end. If the preguideline trajectory had been maintained without the observed change after guideline release, the probability of guideline-concordant prescribing would have been 0.31 (95% CI, 0.15-0.47; P = .001) at study end. In non-children’s hospitals, the probability of guideline-concordant prescribing increased from 0.06 (95% CI, 0.04-0.08) immediately before guideline release to 0.27 (95% CI, 0.20-0.35) at study end. If the preguideline trajectory had continued, prescribing would have been 0.08 (95% CI, 0.01-0.14; P = .004) at study end. Postguideline trajectories were similar between hospital types (difference in probability from beginning to end of final study year: 0.08 absolute increase [95% CI, 0.05-0.10] at children's hospitals vs 0.07 absolute increase [95% CI, 0.04-0.10] at non-children's hospitals; P = .56).

Figure. Guideline-Concordant Prescribing for Pediatric Community-Acquired Pneumonia at Children’s and Non-Children’s Hospitals.

Raw and modeled probabilities of receipt of guideline-concordant therapy (penicillin, amoxicillin, or ampicillin) during hospitalization for uncomplicated pediatric community-acquired pneumonia at children’s and non-children’s hospitals between January 1, 2009, and September 30, 2015, before (quarters −11 through −1) and after (quarters 1 through 15) guideline publication. The dotted line indicates national guideline publication in October 2011.

Discussion

Four years after publication of national pediatric CAP guidelines, only 27% of children admitted to non-children’s hospitals received guideline-concordant therapy compared with 61% in children’s hospitals. This gap is concerning because approximately 70% of children hospitalized with pneumonia receive care in non-children’s hospitals.⁶

The reason for this difference in guideline-concordant prescribing is unknown. It is unlikely attributable to differences in patient populations because we included only healthy children with uncomplicated CAP and adjusted for potential confounders. Studies³ in children’s hospitals have suggested that local implementation efforts may be important in facilitating guideline uptake. Non-children’s hospitals likely have fewer resources to lead pediatric-specific efforts, and care may be influenced by adult CAP guidelines.

Our study is limited by the use of administrative data, potential unknown discrepancies between the databases, and absence of a children’s hospital indicator in Premier Perspectives. However, we identified patients with a validated algorithm, and our results were consistent with prior work in both databases.^3,6 Although guideline-concordant prescribing has increased in both children's and nonchildren's hospitals, non-children’s hospitals appear to be lagging markedly behind children’s hospitals. This discrepancy may represent an important target for antimicrobial stewardship efforts.