Cumulative incidence of discharge according to antibiotic treatment in a cohort of children matched 1:1 according to their propensity to receive β-lactam plus macrolide combination therapy, conditional on baseline demographic, clinical, and radiographic characteristics.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Williams DJ, Edwards KM, Self WH, et al. Effectiveness of β-Lactam Monotherapy vs Macrolide Combination Therapy for Children Hospitalized With Pneumonia. JAMA Pediatr. 2017;171(12):1184–1191. doi:10.1001/jamapediatrics.2017.3225
Is treatment with a β-lactam antibiotic in combination with a macrolide more effective than β-lactam monotherapy among children hospitalized with pneumonia?
In this prospective cohort of more than 1400 children hospitalized with pneumonia, there were no significant differences in length of stay, intensive care admission, rehospitalizations, or recovery at follow-up among children receiving β-lactam plus macrolide combination therapy compared with β-lactam monotherapy.
The results of this study question the routine empirical use of macrolide combination therapy in this population.
β-Lactam monotherapy and β-lactam plus macrolide combination therapy are both common empirical treatment strategies for children hospitalized with pneumonia, but few studies have evaluated the effectiveness of these 2 treatment approaches.
To compare the effectiveness of β-lactam monotherapy vs β-lactam plus macrolide combination therapy among a cohort of children hospitalized with pneumonia.
Design, Setting, and Participants
We analyzed data from the Etiology of Pneumonia in the Community Study, a multicenter, prospective, population-based study of community-acquired pneumonia hospitalizations conducted from January 1, 2010, to June 30, 2012, in 3 children’s hospitals in Nashville, Tennessee; Memphis, Tennessee; and Salt Lake City, Utah. The study included all children (up to 18 years of age) who were hospitalized with radiographically confirmed pneumonia and who received β-lactam monotherapy or β-lactam plus macrolide combination therapy. Data analysis was completed in April 2017.
Main Outcomes and Measures
We defined the referent as β-lactam monotherapy, including exclusive use of an oral or parenteral second- or third-generation cephalosporin, penicillin, ampicillin, ampicillin-sulbactam, amoxicillin, or amoxicillin-clavulanate. Use of a β-lactam plus an oral or parenteral macrolide (azithromycin or clarithromycin) served as the comparison group. We modeled the association between these groups and patients’ length of stay using multivariable Cox proportional hazards regression. Covariates included demographic, clinical, and radiographic variables. We further evaluated length of stay in a cohort matched by propensity to receive combination therapy. Logistic regression was used to evaluate secondary outcomes in the unmatched cohort, including intensive care admission, rehospitalizations, and self-reported recovery at follow-up.
Our study included 1418 children (693 girls and 725 boys) with a median age of 27 months (interquartile range, 12-69 months). This cohort was 60.1% of the 2358 children enrolled in the Etiology of Pneumonia in the Community Study with radiographically confirmed pneumonia in the study period; 1019 (71.9%) received β-lactam monotherapy and 399 (28.1%) received β-lactam plus macrolide combination therapy. In the unmatched cohort, there was no statistically significant difference in length of hospital stay between children receiving β-lactam monotherapy and combination therapy (median, 55 vs 59 hours; adjusted hazard ratio, 0.87; 95% CI, 0.74-1.01). The propensity-matched cohort (n = 560, 39.5%) showed similar results. There were also no significant differences between treatment groups for the secondary outcomes.
Conclusions and Relevance
Empirical macrolide combination therapy conferred no benefit over β-lactam monotherapy for children hospitalized with community-acquired pneumonia. The results of this study elicit questions about the routine empirical use of macrolide combination therapy in this population.
Pneumonia is one of the most common serious infections in childhood, ranking among the top 3 reasons for pediatric hospitalization in the United States annually.1 Pneumonia also accounts for more days of antibiotic use in US children’s hospitals than any other condition,2 making it an important target for antibiotic stewardship efforts. The Pediatric Infectious Diseases Society/Infectious Diseases Society of America national consensus guideline for the management of pneumonia in children recommends narrow-spectrum β-lactam therapy (eg, ampicillin or amoxicillin) for most children with suspected bacterial pneumonia in both inpatient and outpatient settings.3 While β-lactams are highly effective against most common bacterial pneumonia pathogens, including Streptococcus pneumoniae, they do not possess activity against atypical bacteria, such as Mycoplasma pneumoniae, that commonly cause pneumonia in school-aged children and young adults.4,5
Macrolide antibiotics possess in vitro activity against M pneumoniae and Chlamydophila pneumoniae, and the guideline recommends their use when such pathogens are suspected.3 However, with few clinical studies demonstrating the effectiveness of macrolides in children, the guideline grades this recommendation as weak. Recent meta-analyses, including both randomized and nonrandomized studies, noted insufficient evidence to support or refute the use of antibiotics for M pneumoniae–related lower respiratory tract infections in children and adults.6-8 Despite this inadequacy in the study of pediatric pneumonia, the anti-inflammatory properties of macrolides suggest their benefits may also extend beyond direct antibacterial effects.9 Regardless, macrolides are often used as empirical therapy for pneumonia in children.10-12 Thus, determining the potential benefits of macrolide therapy in combination with a β-lactam for the management of childhood pneumonia remains important.
Using data prospectively collected from a cohort of more than 1400 children hospitalized with radiographically confirmed, community-acquired pneumonia, we compared the effectiveness of β-lactam monotherapy with β-lactam plus macrolide combination therapy.
This study was nested within the Centers for Disease Control and Prevention (CDC) Etiology of Pneumonia in the Community (EPIC) Study, a prospective, population-based, active surveillance study of community-acquired pneumonia hospitalizations among children (age, <18 years) conducted between January 1, 2010, and June 30, 2012, at 3 children’s hospitals in Nashville, Tennessee; Memphis, Tennessee; and Salt Lake City, Utah.4 Children were enrolled in the EPIC study if they were hospitalized with signs or symptoms of acute infection (eg, fever), acute respiratory illness (eg, cough), and radiographic evidence of pneumonia. Children with recent hospitalization, severe immunosuppression, cystic fibrosis, tracheostomy, or a clear alternative diagnosis were excluded.
Our research team devised inclusion criteria additional to those of the EPIC study covering empiric antibiotic use. Empirical antibiotic use was classified according to the antibiotics received during the first 2 calendar days of hospitalization. β-Lactams included oral or parenteral second- or third-generation cephalosporins (excluding anti-pseudomonal cephalosporins), as well as penicillin, ampicillin, ampicillin-sulbactam, amoxicillin, and amoxicillin-clavulanate. β-Lactam monotherapy was defined as the exclusive use of 1 or more of these antibiotics. Use of a β-lactam plus an oral or parenteral macrolide (azithromycin or clarithromycin) during the first 2 calendar days of hospitalization served as the macrolide combination therapy group. Children who did not receive antibiotics or who received other classes of antibiotics during the first 2 calendar days of hospitalization were excluded.
EPIC study researchers conducted caregiver interviews and detailed reviews of medical records for all enrolled children. Etiologic assessments included bacterial culture of blood samples, pneumococcal and group A streptococcal polymerase chain reaction, blood tests for 8 respiratory viruses, and nasopharyngeal or oropharyngeal swabs for polymerase chain reaction for 13 respiratory viruses, including M pneumoniae and C pneumoniae. A board-certified pediatric study radiologist, blinded to demographic and clinical information, completed standardized radiographic interpretation at each study hospital. Results of these assessments were not shared with treating clinicians. Caregivers were interviewed 3 to 10 weeks following hospital discharge to gather data regarding self-reported illness recovery and rehospitalizations.
The study was approved by the institutional review board at Vanderbilt University Medical Center, University of Utah, and University of Tennessee Health Sciences Center, plus the CDC. The parents and legal guardians of enrolled children provided written informed consent to the study. Where applicable, age-appropriate assent was also obtained from child participants.
We modeled the association between the empirical antibiotic groups (β-lactam monotherapy vs β-lactam plus macrolide combination therapy) and the length of stay (measured in hours), using multivariable Cox proportional hazards regression. In this analysis, a hazard ratio less than 1 indicates a lower hazard (rate) of hospital discharge for children receiving β-lactam plus macrolide combination therapy compared with children receiving β-lactam monotherapy. There were no deaths or censoring events.
The multivariable analysis included the following covariates, selected a priori and collected via medical records review: age, sex, race/ethnicity, government insurance status, prematurity at birth (if younger than 24 months), chronic comorbidities (grouped as pulmonary, neurologic, cardiovascular, genetic/metabolic, and other [endocrine, renal, hepatic, hematologic, and immunologic]), prior use of antibiotics, clinical examination on admission (presence of altered mental status, chest indrawing, and/or wheezing), admission to intensive care during the first 2 calendar days of hospitalization, and receipt of invasive mechanical ventilation during the first 2 calendar days of hospitalization.
In addition to a confirmation of diagnosis, assessment by a pediatric radiologist noted radiographic features (eg, infiltrate pattern and parapneumonic effusion). These were also included as covariates.
Finally, vital signs on admission were selected as covariates. These included temperature, heart rate, and respiratory rate. In addition, we collected the ratio of oxygen saturation to fraction of inspired oxygen (SpO2/FiO2), and used this to calculate the ratio of partial pressure arterial oxygen and fraction of inspired oxygen (the Pao2/FiO2 ratio, or PF ratio), as described in previous studies.13
Owing to age-based differences in heart and respiratory rates, interaction terms between each of these variables and age were also included. Analyses accounted for the clustering of observations at the individual sites. Adjusted hazard ratios (aHRs) and 95% confidence intervals (95% CIs) were reported.
Secondary analyses were performed using propensity score matching and weighting to further address the possibility of residual confounding . Covariates from the primary analysis were used to calculate a propensity score, using a logistic regression model with antibiotic therapy (β-lactam monotherapy vs β-lactam plus macrolide combination therapy) as the dependent variable. Observations from exposure groups were matched 1:1 on propensity scores, using matching without replacement and a caliper width equal to 0.25 times the standard deviation of the linear predictor of the propensity score.14 Visual inspection of the distribution of propensity scores between the matched exposure groups demonstrated good overlap, and standardized differences were calculated to verify postmatching balance.
A second propensity score–matched analysis was also conducted to stratify matching by study site. In this analysis, residual imbalance was noted for respiratory rate at one study site; thus, respiratory rate was included as a covariate in this model.
Finally, since propensity score matching reduced the effective sample size, we also conducted a propensity scores weighted analysis. This approach achieves balance in the distribution of covariates by weighting the study observations into a well-balanced synthetic study population, and typically retains more observations than propensity score matching.15 Stabilized weights were calculated, and extreme weights were truncated at the fifth percentiles to limit their influence. From the propensity score model, we derived the reciprocal of the probability of receiving combination therapy; we used this to calculate inverse weights for the probability of treatment. A weighted Cox proportional hazards regression model with robust standard errors was used to assess the hospital discharge rate of the 2 exposure groups.15,16
Several planned subgroup analyses were conducted among children, with and without selected characteristics that might modify the impact of antibiotic selection on length of stay. This included age 5 years and older, detection of atypical bacteria (M pneumoniae and C pneumoniae), wheezing, and admission to intensive care. Analyses stratified by hospital were also conducted. For each subgroup, a new propensity score was calculated. Then, propensity score deciles were included as the sole covariate in a Cox proportional hazards regression model.
Finally, we used propensity score–adjusted logistic regression models, adjusted with propensity score deciles, to evaluate intensive care admission occurring after the first hospital day and, for those with follow-up data available, recovery and rehospitalization. The last 2 of these 3 secondary outcomes were based on self-report at the time of follow-up. All analyses were conducted using Stata 13.1 (StataCorp LLC).
Among the 2358 children with radiographically confirmed pneumonia who enrolled in the EPIC study, 371 children (15.7%) did not receive antibiotics within the first 2 calendar days of hospitalization, and 569 (24.1%) received antibiotics other than β-lactams (with or without macrolides). These children were excluded from this study (Figure 1). The remaining 1418 children (60.1%) included 1019 (71.9%) who received β-lactam monotherapy and 399 (28.1%; range across hospitals 22.6%-32.6%) who received a β-lactam plus a macrolide; these 1418 patients constituted the study population (Table 1). Of the children who received a macrolide, 392 (98.2%) received azithromycin. The study population included 1045 children (73.7%) who had a virus detected with or without bacterial co-detection; 65 (4.6%) had both a virus and bacteria detected. An atypical bacteria was detected in an additional 125 children (8.8%), and of these, M pneumoniae accounted for 119 (95.2%). Forty-four patients (3.1%) had another bacteria detected, and 235 (17.5%) had no pathogen detected.
The propensity score–matched cohort retained 554 children (39.1%) (Table 1). Important differences in demographic (eg, age), clinical (eg, chronic comorbidities and chest indrawing), and radiographic (eg, infiltrate pattern and pleural effusion) characteristics were noted between groups in the unmatched cohort. These differences were no longer present in the propensity score–matched cohort.
In the unmatched cohort, the median length of stay for children receiving β-lactam monotherapy was 55 hours (interquartile range [IQR], 39-91 hours) and 59 hours (IQR, 41-89 hours) for those receiving a β-lactam in combination with a macrolide (unadjusted HR, 1.01; 95% CI, 0.90-1.14) (Table 2). The multivariable analysis demonstrated no significant difference in length of stay between the groups (adjusted HR, 0.87; 95% CI, 0.74-1.01). In the propensity score–matched analysis, the median length of stay was 53 hours (IQR, 36-85 hours) for children receiving β-lactam monotherapy and 62 hours (IQR, 42-91 hours) for those receiving β-lactams in combination with a macrolide (HR, 0.88; 95% CI, 0.74-1.03) (Figure 2). Results for stratified, propensity-matched and propensity-weighted secondary analyses were similar (Table 2).
Our subgroup analyses compared length of stay between empirical antibiotic regimens across 4 groups in whom macrolide combination therapy might offer the most benefit. These included children older than 5 years (n = 406, of whom 224, or 55.2%, received combination therapy), children with atypical bacteria detected (n = 125, of whom 76, or 60.8%, received combination therapy), children admitted to intensive care (n = 215, of whom 63, or 29.3%, received combination therapy), and children with acute wheezing (n = 603, of whom 155, or 25.7%, received combination therapy). In analyses stratified by site, there was no significant difference in length of stay between group members who had received β-lactam monotherapy and those receiving β-lactams plus a macrolide (Table 3).
In total, 227 children (16.0%) were admitted to intensive care. Of these, 82 (36.1%) were transferred to intensive care after the first hospital day. In this group, there was no significant difference between those receiving β-lactam monotherapy (n = 57; 35%) and those receiving a β-lactam plus a macrolide (n = 25; 38%; adjusted odds ratio [aOR], 1.04; 95% CI, 0.41-2.64). There were no in-hospital deaths.
Follow-up data were available for 873 children (61.6%), including 616 children from the unmatched cohort who received β-lactam monotherapy (n = 616/873; 70.5%) and 257 who received combination therapy (n = 257/873; 29.4%). Overall, 769 children (n = 769/873; 88.1%) had recovered from their initial illness at the time of follow-up, including 535 children (n = 535/616; 86.9%) who received β-lactam monotherapy and 234 (n = 234/257; 91.1%) who received combination therapy (aOR, 1.11; 95% CI, 0.59-2.07). Thirty-three children (33/616 = 5.4%) receiving monotherapy and 5 (5/257 = 2.0%) receiving combination therapy were rehospitalized (aOR, \0.45; 95% CI, 0.15-1.35).
In our multicenter, prospective observational study conducted among 1418 children hospitalized with radiographically confirmed community-acquired pneumonia, empirical therapy with a macrolide in combination with a β-lactam conferred no benefit compared with β-lactam monotherapy. Findings from propensity score–matched and propensity score–weighted analyses were essentially identical to those in the unmatched cohort. Results were also similar among children older than 5 years, those with atypical pathogens detected, those admitted to intensive care, and those with acute wheezing. Similar proportions of children treated with either of the 2 empirical antibiotic regimens were transferred to intensive care after the first hospital day, and nearly all children in both groups had recovered in the 3 to 10 weeks following their hospital discharge.
Two prior observational studies that compared β-lactam monotherapy vs β-lactam plus macrolide combination therapy for children hospitalized with pneumonia reported a shorter length of stay for children receiving combination therapy.17,18 However, those studies relied on administrative data and systematically excluded infants, children with comorbid conditions, and those with severe illness. In contrast, our study used data from a prospective cohort using a strict definition of clinical and radiographically confirmed pneumonia, and adjusted for potential confounding by including detailed etiologic assessments and important covariates unavailable in administrative data sources (eg, vital signs, chest indrawing, and infiltrate patterns). It is therefore difficult to directly compare these studies.
A strength of our study was that it was nested within a larger pneumonia etiology study that included comprehensive and systematic identification of radiographically confirmed pneumonia and extensive microbiologic assessments for viruses and bacteria. (More than 70% of included children had a virus detected; fewer than 10% had atypical bacteria detected, and typical bacteria were even less commonly detected.)
Nevertheless, we acknowledge that it is very difficult to determine the cause of pneumonia using current diagnostics and without direct sampling of the lung. If the etiologies of pneumonia included in our study were primarily a mixture of viruses and bacteria that responded to β-lactams, then it may not be altogether surprising that the 2 empirical treatment regimens yielded similar outcomes. In addition, others have suggested that macrolides might provide benefit to those with viral pneumonia due to their immunomodulatory effects, although this hypothesis is not supported by our data. But it is possible that the antibacterial and anti-inflammatory properties of macrolides might benefit some children with pneumonia, such as those with underlying chronic lung diseases or those with very severe pneumonia.19,20 Our subgroup analyses focused on several of these populations, but did not demonstrate a clear benefit of empirical β-lactam plus macrolide combination therapy. Importantly, the frequency of very severe pneumonia and chronic lung diseases other than asthma was low in our study, and analyses of these groups might have been underpowered to detect important treatment differences. Thus, confirming these findings in a large, multicenter study focused on children most likely to benefit from empirical macrolide combination therapy will be important.
Judicious antibiotic selection is critical to slowing the progression of antimicrobial resistance, and excessive use of macrolides has been an important target.21-23 Despite overall declines in antibiotic use in US outpatient children with acute respiratory illness between 2000 and 2010, the use of broad-spectrum antibiotics in this same population nearly doubled, largely as a result of increased use of macrolides.24 A national study of emergency department visits for childhood pneumonia25 conducted prior to the release of the PIDS/IDSA guideline3 found that macrolides were the most common antibiotics prescribed and accounted for nearly half of all antibiotics given in visits that resulted in such prescriptions. A similar study conducted among 29 US children’s hospitals revealed that 35% of children hospitalized with pneumonia received a macrolide.12 Two prior studies, including one using data from the EPIC study, showed that publication of the guideline was associated with only modest declines in macrolide use for children hospitalized with pneumonia.26,27
In this study, almost 30% of children received macrolide combination therapy, even though atypical pathogens were detected in less than 9%. While detection of an atypical pathogen was more common among those receiving a macrolide, nearly 40% of children with a detected atypical pathogen did not receive a macrolide. This discrepancy underscores the challenge of empirical macrolide use for pediatric pneumonia. While the PIDS/IDSA guideline recommends consideration of atypical coverage when such pathogens are suspected,3 there are no clinical or radiographic criteria that reliably distinguish atypical pathogens from viral or other bacterial causes of pneumonia. The notable exception is age, as atypical pathogens only rarely cause pneumonia in preschool-aged children or younger.5 In older children, the increasing availability of rapid and sensitive molecular diagnostics for M pneumoniae offers perhaps the most promising strategy, reserving treatment for children with a positive test result.
Even so, our study did not demonstrate benefits of empirical macrolide therapy in those with atypical bacteria. Two recent meta-analyses examining macrolide therapy for children with M pneumoniae pneumonia reached similar conclusions.6,7 Thus, clinicians must weigh the theoretical individual benefits of empirical macrolide therapy against the risk of adverse drug effects and the societal risks associated with antimicrobial resistance.10,11,28
The nonrandomized, observational design of our study is a limitation. Baseline differences between those receiving β-lactam monotherapy and β-lactam plus macrolide combination therapy raise the potential for confounding. To minimize this, we used multivariable regression modeling to adjust for important covariates and created a propensity score–matched cohort to further address confounding. Results were essentially identical to the unmatched cohort.
Another potential limitation is that empirical antibiotic exposure was defined based on therapy received during the first 2 hospital days, and subsequent antibiotic use or duration of therapy was not evaluated. It is possible that therapy was broadened or narrowed after the second hospital day. However, given that the median length of stay was fewer than 3 days, any misclassifications would likely be minimal.
Similarly, we did not assess treatment compliance after discharge, which may have affected our secondary postdischarge outcomes. The limited follow-up data available precluded more robust assessments of these secondary outcomes. Finally, given the short length of stay for most children, it is also possible that combination therapy conferred benefits not captured in our primary or secondary outcome assessments, such as fewer symptoms or a quicker return to normal activity.
In summary, in our study of 1418 children hospitalized with community-acquired pneumonia, adding a macrolide to empirical β-lactam therapy conferred no treatment benefit over β-lactam monotherapy. These results were consistently observed in several populations in whom macrolide therapy are presumed to be most beneficial, including those with atypical pathogens detected. Our study questions routine use of empirical macrolide combination therapy in children hospitalized with pneumonia and represents an important potential target for antibiotic stewardship.
Corresponding Author: Derek J. Williams, MD, MPH, Vanderbilt University Medical Center, CCC5324 Medical Center North, 1161 21st Ave S Nashville, TN 37232 (firstname.lastname@example.org).
Accepted for Publication: July 26, 2017.
Published Online: October 30, 2017. doi:10.1001/jamapediatrics.2017.3225
Author Contributions: Dr Williams had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Williams, Edwards, Self, Zhu, Arnold, Ampofo, Pavia, Hicks, Jain, Grijalva.
Acquisition, analysis, or interpretation of data: Williams, Edwards, Self, Zhu, Arnold, McCullers, Pavia, Anderson, Hicks, Bramley, Jain, Grijalva.
Drafting of the manuscript: Williams, Anderson, Bramley.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Williams, Self, Zhu, Anderson, Grijalva.
Obtained funding: Williams, Edwards, McCullers, Pavia, Jain.
Administrative, technical, or material support: Williams, Edwards, Self, McCullers, Pavia, Bramley, Jain.
Study supervision: Edwards, McCullers, Ampofo, Pavia, Jain, Grijalva.
Conflict of Interest Disclosures: Dr Self reports payment for participation in a scientific advisory board meeting for Cempra Pharmaceuticals. Dr Anderson reports payment from AbbVie for consulting and research funding from Regeneron and MedImmune. Dr Pavia reports payment for participation on data safety and monitoring boards for National Institutes for Health, Alios Pharmaceuticals, and Janssen Pharmaceuticals. No other disclosures are reported.
Funding/Support: This work was supported by the National Institute of Allergy and Infectious Diseases of the National Institutes of Health under award K23AI104779 (Dr Williams), the National Institute of General Medical Sciences under award K23GM110469 (Dr Self), and the Agency for Healthcare Research and Quality under award R03HS022342 (Dr Grijalva). The EPIC study was supported by the Influenza Division in the National Center for Immunizations and Respiratory Diseases at the CDC through cooperative agreements with each study site and was based on a competitive research funding opportunity.
Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the National Institute of Allergy and Infectious Diseases, the National Institute of General Medical Sciences, the Agency for Healthcare Research and Quality, or the CDC.
Create a personal account or sign in to: