Postdischarge treatment of acute osteomyelitis in children requires weeks of antibiotic therapy, which can be administered orally or intravenously via a peripherally inserted central catheter (PICC). The catheters carry a risk for serious complications, but limited evidence exists on the effectiveness of oral therapy.
To compare the effectiveness and adverse outcomes of postdischarge antibiotic therapy administered via the PICC or the oral route.
Design, Setting, and Participants
We performed a retrospective cohort study comparing PICC and oral therapy for the treatment of acute osteomyelitis. Among children hospitalized from January 1, 2009, through December 31, 2012, at 36 participating children’s hospitals, we used discharge codes to identify potentially eligible participants. Results of medical record review confirmed eligibility and defined treatment group allocation and study outcomes. We used within- and across-hospital propensity score–based full matching to adjust for confounding by indication.
Postdischarge administration of antibiotics via the PICC or the oral route.
Main Outcomes and Measures
The primary outcome was treatment failure. Secondary outcomes included adverse drug reaction, PICC line complication, and a composite of all 3 end points.
Among 2060 children and adolescents (hereinafter referred to as children) with osteomyelitis, 1005 received oral antibiotics at discharge, whereas 1055 received PICC-administered antibiotics. The proportion of children treated via the PICC route varied across hospitals from 0 to 100%. In the across-hospital (risk difference, 0.3% [95% CI, −0.1% to 2.5%]) and within-hospital (risk difference, 0.6% [95% CI, −0.2% to 3.0%]) matched analyses, children treated with antibiotics via the oral route (reference group) did not experience more treatment failures than those treated with antibiotics via the PICC route. Rates of adverse drug reaction were low (<4% in both groups) but slightly greater in the PICC group in across-hospital (risk difference, 1.7% [95% CI, 0.1%-3.3%]) and within-hospital (risk difference, 2.1% [95% CI, 0.3%-3.8%]) matched analyses. Among the children in the PICC group, 158 (15.0%) had a PICC complication that required an emergency department visit (n = 96), a rehospitalization (n = 38), or both (n = 24). As a result, the PICC group had a much higher risk of requiring a return visit to the emergency department or for hospitalization for any adverse outcome in across-hospital (risk difference, 14.6% [95% CI, 11.3%-17.9%]) and within-hospital (risk difference, 14.0% [95% CI, 10.5%-17.6%]) matched analyses.
Conclusions and Relevance
Given the magnitude and seriousness of PICC complications, clinicians should reconsider the practice of treating otherwise healthy children with acute osteomyelitis with prolonged intravenous antibiotics after hospital discharge when an equally effective oral alternative exists.
Osteomyelitis is a serious bacterial infection that requires a prolonged course of antibiotic therapy to minimize the risk for treatment failure. After resolution of acute symptoms (eg, fever, pain, and disability) in the hospital, most children are discharged to complete a 4- to 6-week course of antibiotic therapy at home.1 Until recently, home antibiotic therapy was commonly delivered intravenously via a central venous catheter (usually a peripherally inserted central catheter [PICC]), a practice known as outpatient parenteral antibiotic therapy.2 However, in 1997, Peltola et al3 published a case series of children in Finland with Staphylococcus aureus osteomyelitis that showed excellent outcomes with early transition to oral antibiotic therapy after normalization of the C-reactive protein level.
The route of administration chosen for outpatient antibiotic therapy has obvious implications for the overall experience of the child and his or her caregivers. The PICCs are effective for delivering high concentrations of antibiotic but are fraught with infectious, thrombotic, and mechanical complications.4 Oral antibiotics with high bioavailability make oral step-down therapy an appealing alternative, but only small case series3,5,6 and no clinical trials document their effectiveness. A prior study7 that used only administrative data in the Pediatric Health Information System (PHIS) to compare the oral and PICC routes found no difference in the rates of treatment failure, but several important missing elements may have limited the validity and interpretation of those results, including the validation of the osteomyelitis diagnosis and treatment choice, adjustment for the severity of illness (eg, drainage procedures during the index hospitalization), and information about the reasons for readmissions and revisits. These study design limitations may explain in part why most hospitals subsequently did not reduce their use of the PICC route in favor of the less invasive oral route.
To supplement the PHIS administrative data with additional clinical information obtained from detailed review of medical records, we sought to compare the effectiveness of oral antibiotics with intravenous (IV) antibiotics delivered via a PICC in children who received prolonged home antibiotic therapy after hospitalization for osteomyelitis. We tested the following hypotheses: (1) children who receive oral antibiotic therapy at home would not have more treatment failures than comparable children who receive IV antibiotic therapy at home; and (2) children receiving IV antibiotic therapy at home would have more emergency department (ED) visits and hospitalizations owing to complications with their PICC line.
The study protocol was approved by the institutional review boards of all participating sites. Informed consent was waived. We used the PHIS database to identify a retrospective cohort of potentially eligible children and adolescents (hereinafter referred to as children) at 38 children’s hospitals from January 1, 2009, through December 31, 2012. As reported previously, the PHIS database contains detailed hospital administrative and billing data from freestanding children’s hospitals affiliated with the Children’s Hospital Association, Overland Park, Kansas.8 We validated and supplemented the PHIS data for each child by having local hospital physicians and trained research assistants perform detailed reviews of the medical records to (1) confirm eligibility; (2) determine postdischarge antibiotic choice, route, and duration of therapy; (3) review all return visits to the ED and all hospitalizations within 6 months for evidence of treatment failure, adverse drug reaction, or PICC line complication; and (4) extract the findings of cultures of blood, bone, and joint aspiration fluid (ie, causative organism and susceptibilities). These data were entered and stored in a REDCap (Research Electronic Data Capture) database and subsequently merged with the administrative and the billing data obtained from PHIS.
Quiz Ref IDChildren were included if they were at least 2 months but younger than 18 years on the date of admission and were discharged from January 1, 2009, through December 31, 2012, with a discharge code from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) for acute osteomyelitis (730.01-730.09) or unspecified osteomyelitis (730.2-730.29) in any 1 of 21 discharge diagnosis fields (Figure 1).
Quiz Ref IDWe excluded children who (1) were hospitalized in the 6 months preceding the index admission with an ICD-9-CM discharge diagnosis of acute, unspecified, or chronic osteomyelitis; (2) had a concurrent or a previous ICD-9-CM code for a chronic cardiac, hematologic, immunologic, oncologic, or respiratory condition that would increase the child’s risk for treatment failure with either treatment modality (eTable 1 in the Supplement)9; (3) were not admitted to the hospital through the ED (thus excluding children transferred from another hospital ward at some unknown point in their disease course); (4) were transferred to another facility; (5) had other specified sites of osteomyelitis (ICD-9-CM code 730.08); or (6) had a length of stay of less than 2 or more than 14 days.
In addition, 2 investigators (R.K. and R.S.) independently reviewed the primary and secondary discharge diagnosis and procedure codes for the index admissions of all the children. They excluded all children who had codes consistent with the inability to take antibiotics orally or via a nasogastric or a gastrostomy tube, problems absorbing antibiotics enterally, or acquired or congenital immunodeficiencies considered a high risk for treatment failure if the oral route is in fact inferior; who were medically fragile and would not be exposed to the risks of a potentially inferior treatment route; with cellulitis or septic arthritis listed as the primary discharge diagnosis; with orthopedic hardware indicating an osteomyelitis secondary to trauma or a foreign body; with bone fractures, indicating osteomyelitis secondary to trauma; with conditions associated with immobilization and the development of pressure ulcers and difficult-to-treat osteomyelitis; and with osteomyelitis of the head, the face, and orbits (eTable 2 in the Supplement).
The clinical sites were provided with lists of patients and hospitalizations for the review of the medical records based on the eligibility criteria applied to the PHIS data. These criteria included the index hospitalization and all return visits (ED or hospitalization) as long as 6 months later. Any children who were found to meet any of the aforementioned exclusion criteria were excluded at the time of the medical record review.
We used information from the medical record review to classify children into 1 of 2 possible treatment groups at the time of discharge from the hospital: home IV antibiotic therapy via a PICC or home oral antibiotic therapy. Children who had a PICC line placed during the index hospitalization but removed before discharge and were prescribed oral antibiotics were classified as receiving oral therapy.
All outcomes were determined using information obtained from the medical record review. The primary outcome for the study was treatment failure, defined as revisit to the ED or a rehospitalization for a change in the antibiotic prescribed or its dosage, prolongation of antibiotic therapy, conversion from the oral to the PICC route, bone abscess drainage, debridement of necrotic bone, bone biopsy, drainage of an abscess of the skin or muscle, arthrocentesis, or diagnosis of a pathologic fracture. Quiz Ref IDSecondary outcomes included a return to the ED or a rehospitalization for (1) an adverse drug reaction (defined as vomiting and/or diarrhea, dehydration, Clostridium difficile infection, allergic reaction, urticaria, anaphylaxis, drug-induced neutropenia, acute kidney injury, Stevens-Johnson syndrome, erythema multiforme, or other); (2) a PICC complication (defined as fever evaluation, infection at the site of the PICC insertion, blood stream infection, sepsis, and thrombosis, breakage, repair, adjustment, manipulation, or removal of the PICC line with or without insertion of a new line); and (3) all the aforementioned adverse outcomes for a composite of treatment failure, adverse drug reaction, and PICC line complication.
The PHIS database provided information on age, sex, race, ethnicity, payer, anatomic location of the infection (based on ICD-9-CM fifth-digit subclassification), and performance of surgical procedures during the index hospitalization (ICD-9-CM procedure codes for arthrocentesis, arthrotomy, osteotomy, soft-tissue incision, and drainage). The review of the medical records provided the findings of cultures of blood, bone, and joint aspiration fluid (ie, the causative organism and susceptibilities).
To adjust for potential confounding, we implemented propensity score–based full matching10,11 to balance patient-level covariates in the PICC and oral groups. Full matching links each child who received antibiotics via the PICC route to the most similar child receiving antibiotics via the oral route or vice versa in a manner that results in varying numbers of matches in each resulting set. To estimate propensity scores, we used a logistic regression with the following patient-level covariates: age (≤5 vs >5 years), race (white vs nonwhite), insurance (government vs commercial/self-pay), length of stay (in days), location of the infection (shoulder, arm, and hand; pelvis and thigh; lower leg and foot; multiple sites; and unspecified), 4 indicators for a surgical procedure (arthrocentesis, osteotomy, soft-tissue incision and drainage, and arthrotomy), and isolation of the causative pathogens in the cultures of blood, bone, and joint aspiration fluid (findings negative or positive for methicillin-resistant S aureus [MRSA], methicillin-sensitive S aureus, or other organisms).
Patients were matched to balance the patient-level confounders across and within hospitals. Matching across hospitals attempts to find the best matches based on patient-level factors under the assumption that patients across different hospitals are the same if they share common observed and recorded characteristics. This method ignores the possibility of confounding by hospital (outcomes vary by hospital in ways that cannot be explained by observed and recorded patient characteristics) but attempts to address the issues of possible confounding by indication (patients who appear to be the same are selected for the PICC route because they are sicker or the PICC route is indicated for reasons not disclosed by observed patient characteristics). Matching within hospitals or hospital groups mimics a typical stratified randomized clinical trial in that patients who share the same characteristics within the same hospital are compared.12 These within-hospital contrasts control for confounding by hospital (by unobserved, hospital-specific patient or treatment characteristics) but might be more susceptible to confounding by indication. By using both approaches, we examined the robustness of model selection to hospital- and patient-level unobserved confounding. Propensity score modeling and full matching were performed using the program Optmatch (R software package, version 3.1).10
For the response models, we used stratified approaches and marginal models. The stratified approaches imposed conditions on the matched set. These models dropped patients if a matched set did not contain patients with differing outcomes. By contrast, the marginal models used all patients without stratification with the justification that the matching and the weighting resulted in 2 groups of patients who were similar except for the type of antibiotic therapy prescribed on discharge.
We implemented several response models and used resampling to check the 95% CIs.11,13 Conditional logistic regression was the primary method for estimating ratios (for the association of the treatment route and outcomes). The treatment route was the only predictor in these models. We used fixed-effects linear regression (the first differencing method) to estimate risk differences of the outcomes between the patients receiving therapy via the PICC and the oral routes.14 Alternatively, we used weighted logistic regression to estimate the effect of the PICC route on outcomes among the children with the characteristics of the PICC patients. Weighted models with marginal standardization15 estimated the risks and their differences.
All variance estimates and CIs were confirmed using bootstrap resampling (999 replicates).11 Investigations of effect modification by the child’s age (≤5 vs >5 years) and the culture-verified presence of MRSA on the association of the treatment route and outcomes proceeded in the same fashion by means of propensity score–based full matching. For age and MRSA status, this method developed propensity scores separately within the age and MRSA groups. The process resulted in separate matched sets of patients receiving antibiotics via the PICC and the oral routes for younger and older children (or for patients with and without MRSA). However, we were unable to match within hospital owing to the sparse data. We estimated separate rates of outcome by age group (and MRSA status group) via response models with interaction terms for the PICC route by age (or the PICC route by MRSA status).
For all response models and resampling, we used the programs logit, clogit, and bootstrap with a custom program for this application. All programs are found in a commercially available software package (Stata, version 13.1; StataCorp).
We identified 8555 encounters from 38 hospitals that had discharge ICD-9-CM codes for acute or for unspecified osteomyelitis. A total of 6064 encounters met exclusion criteria derived from the PHIS data. Review of the medical records was conducted on the remaining 2491 encounters, and 431 additional encounters were excluded, among them 68 children from 2 hospitals that did not review any records. Other reasons for exclusion are listed in Figure 1. A total of 2060 children from 36 hospitals were included in the final cohort, nearly evenly divided between children discharged to complete a course of oral antibiotics (n = 1005) vs a course of IV antibiotics delivered via a PICC (n = 1055). The proportion of children discharged to complete antibiotic therapy via the PICC route varied considerably across hospitals (range, 0%-100%), with some hospital representation at every decile of PICC use and no relationship to volume of osteomyelitis cases (Figure 2).
Table 1 describes the clinical and demographic characteristics of the study cohort. Most of the children were male (62.3%), ranged in age from 5 to 13 years (54.1%), were white (62.5%), and were non-Hispanic (71.5%). The median length of stay was 6 (interquartile range, 5-8) days. The most common site of infection was the lower extremity (the lower leg, ankle, and/or foot; the pelvis and thigh) (80.4%). At least 1 culture (blood, bone, or joint aspiration fluid) was positive for a causative organism in 57.4% of children. A minority underwent a surgical procedure such as arthrocentesis (17.1%), osteotomy (39.0%), soft-tissue incision and drainage (13.5%), and arthrotomy (16.7%). The treatment groups were similar in their clinical and demographic characteristics, although the oral therapy group had slightly fewer children younger than 1 year, fewer surgical procedures, and more children who were nonwhite, Hispanic, and government insured. More children in the oral therapy group had culture findings negative for causative organisms.
The most common postdischarge antibiotic therapy prescribed for the oral group included clindamycin palmitate hydrochloride (50.4%) and cephalexin (37%); for the PICC group, clindamycin phosphate (35.5%), cefazolin sodium (33.4%), ceftriaxone sodium (11.4%), and vancomycin hydrochloride (10.1%). Children in the oral therapy and PICC groups were prescribed postdischarge antibiotics for a mean of 32 (interquartile range, 28-37) and 27 (interquartile range, 20-35) days, respectively.
The small differences in the clinical and demographic characteristics between the 2 treatment cohorts were successfully reduced in the across- and within-hospital matched cohorts (Table 2). In unmatched analyses, the rates of treatment failure were nearly identical in the oral therapy (5.0%) and PICC (6.0%) groups. Similarly, in across-hospital (risk difference, 0.3% [95% CI, −0.1% to 2.5%]) and within-hospital (risk difference, 0.6% [95% CI, −0.2% to 3.0%]) matched analyses, children in the oral therapy group (reference group) did not have more treatment failures than those in the PICC group. Rates of adverse drug reaction were low (<4% in both groups) but slightly greater in the PICC group in across-hospital (risk difference, 1.7% [95% CI, 0.1%-3.3%]) and within-hospital (risk difference, 2.1% [95% CI, 0.3%-3.8%]) matched analyses. Of the 1055 children in the PICC group, 158 (15.0%) had a PICC-related complication that required an ED visit (n = 96), a rehospitalization (n = 38), or both (n = 24). As a result, the PICC group had a significantly higher risk of requiring a return ED visit or for rehospitalization for any adverse outcome in across-hospital (risk difference, 14.6% [95% CI, 11.3%-17.9%]) and within-hospital (risk difference, 14.0% [95% CI, 10.5%-17.6%]) matched analyses (Table 3). No deaths occurred in either group. Unstratified results weighted to the number of patients in the PICC group produced similar estimates with somewhat wider 95% CIs. Risk differences from the weighted but unstratified analysis produced estimates similar to the stratified results. eTable 3 in the Supplement provides a complete accounting of the reasons for the primary and the secondary outcomes.
In stratified analyses, we found no clinically relevant difference in the rates of treatment failure for children 5 years and younger (risk difference for across-hospital matched analysis, −1.3% [95% CI, −4.7% to 2.1%]; risk difference for within-hospital matched analysis, −2.8% [95% CI, −7.4% to 1.7%]). However, the risk for treatment failure was increased in children older than 5 years who had received antibiotics via the PICC route (risk difference for across-hospital matched analysis, 3.8% [95% CI, 1.3%-6.3%]; risk difference for within-hospital matched analyses, 4.6% [95% CI, 2.2%-7%]). Tests for the interaction of age group by route of antibiotic at discharge were statistically significant (P = .02 and P = .01 for the across- and within-hospital matching, respectively). Isolation of MRSA as the causative organism did not modify the effect of the treatment route on the outcome of treatment failure.
Quiz Ref IDChildren hospitalized with osteomyelitis and discharged to complete a course of antibiotic therapy via the oral route did not have a higher rate of treatment failure than their peers who received their antibiotic therapy via the PICC route. The lower bounds of the 95% CIs for the increased risk for treatment failure with oral therapy of −0.9% and −1.7% in across- and within-hospital matched analyses support the effectiveness of the oral route for postdischarge antibiotic therapy for osteomyelitis. Quiz Ref IDHowever, the high frequency of PICC-related complications requiring ED revisits and/or rehospitalization means that children treated via the PICC route had a higher risk (by 14 percentage points) for adverse events compared with their propensity-matched peers who received antibiotics via the oral route. Despite concerns about treating younger (≤5 years) children or those with MRSA infections with oral antibiotics, our stratified analyses showed that the treatment failure rates between the oral and the PICC routes were not meaningfully worse in these patient subgroups.
The ideal study design to answer our research question would have been a multicenter randomized clinical trial. However, prior research7 suggests that the decision to deliver postdischarge antibiotic therapy via the PICC route or the oral route might reflect individual hospital culture and/or tradition. Clinicians in hospitals that discharge most of their patients with antibiotic therapy delivered via the oral route might resist randomizing their patients to the more invasive and potentially problematic PICC route, whereas providers in hospitals that use mostly the PICC route might worry about the effectiveness of the oral route. Thus, the lack of hospital-level equipoise about the comparative effectiveness of the 2 treatment options makes the conduct of such a trial challenging and costly. Using an observational study design and controlling for hospital and patient factors, we matched patients in a manner that would mimic a conventional stratified randomized clinical trial. This study design is limited by the sample size in the number of potential confounders for matching, but we believe that we included the key variables that indicate severity of illness, such as the length of stay, the causative organism, the need for a drainage procedure, and location of the infection. We were also limited to matching patients on observed covariates. Other unmeasured confounders might have influenced treatment outcomes.
A prior study of this question,7 which used only administrative data, included a similar number of children hospitalized at 1 of 29 CHA hospitals from 2000 through 2005 who were also evenly divided between the oral and the PICC routes for therapy. The results were virtually identical to our results, with nearly equivalent rates of treatment failure requiring hospitalization in the PICC (5%) and the oral therapy (4%) groups and a rate of PICC-related complications requiring rehospitalization of 3.4% of patients (reliable information on ED revisits was not available). The present study improves on the prior one in that we performed a review of the medical records to validate the exposure and outcome variables, and we used within- and across-hospital matching to account for hospital- and patient-level confounders. The treatment failure and PICC complication rates detected in the present study were slightly higher, possibly reflecting improved ascertainment of outcomes afforded by the review of the medical records. Also, unlike the prior study, which included children hospitalized from 2000 through 2005, the present one included children hospitalized from 2009 through 2012, when culture-proved MRSA was far more prevalent (16.3%) and children were empirically treated for MRSA with clindamycin and/or vancomycin (45%-50%). Thus, the oral route appears to be effective even in the MRSA era.
In this retrospective comparative study of antibiotic therapy delivered via the PICC and the oral routes after hospital discharge in children with acute osteomyelitis, we found no advantage of the more invasive PICC route. Given the magnitude (15.0% of all children in the PICC group) and gravity (ie, bloodstream infection, thromboembolism, and line breakage) of the PICC-related complications, clinicians should reconsider the practice of treating otherwise healthy children with osteomyelitis with prolonged IV therapy when an effective oral alternative exists.
Accepted for Publication: October 9, 2014.
Corresponding Author: Ron Keren, MD, MPH, Division of General Pediatrics, Center for Pediatric Clinical Effectiveness, Children’s Hospital of Philadelphia, Abramson Research Building, Room 1347, Philadelphia, PA 19104 (firstname.lastname@example.org).
Published Online: December 15, 2014. doi:10.1001/jamapediatrics.2014.2822.
Author Contributions: Dr Keren had full access to all 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: Keren, Shah, Srivastava, Rangel, Hartley, Bryan, Hall, Localio, Luan, Parker.
Acquisition, analysis, or interpretation of data: Keren, Shah, Rangel, Bendel-Stenzel, Harik, Hartley, Lopez, Seguias, Tieder, Bryan, Gong, Hall, Localio, Luan, deBerardinis, Parker.
Drafting of the manuscript: Keren, Rangel, Hartley, Lopez, Seguias, Tieder, Bryan, Gong, Hall, Localio, deBerardinis.
Critical revision of the manuscript for important intellectual content: Keren, Shah, Srivastava, Rangel, Bendel-Stenzel, Harik, Hartley, Lopez, Seguias, Bryan, Localio, Luan, Parker.
Statistical analysis: Keren, Harik, Bryan, Gong, Hall, Localio, Luan.
Obtained funding: Keren, Shah, Localio.
Administrative, technical, or material support: Keren, Lopez, Seguias, deBerardinis, Parker.
Study supervision: Keren, Rangel, Hartley, Lopez, Tieder, Bryan, Localio, deBerardinis.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported by the Patient Centered Outcomes Research Institute.
Role of the Funder/Sponsor: The funding source 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.
Group Information: The following Pediatric Research in Inpatient Settings research network collaborators conducted and/or supervised the reviews of the medical records at participating hospitals but did not fulfill authorship criteria: Chris Miller, MD, Primary Children’s Hospital, Salt Lake City, Utah; Wendy Hoffner, MD, Children’s National Medical Center, Washington, DC; Suchitra Rao, MD, Children’s Hospital Colorado, Aurora; Derek J. Williams, MD, MPH, Vanderbilt Children’s Hospital, Nashville, Tennessee; Cynthia D. Cross, MD, Le Bonheur Children’s Hospital, Memphis, Tennessee; Jeffrey Colvin, MD, Children’s Mercy Hospital and Clinics, Kansas City, Missouri; Susan Wu, MD, Children’s Hospital Los Angeles, Los Angeles, California; Waheeda Samady, MD, Lurie Children’s Hospital, Chicago, Illinois; Paul Ishimine, MD, Rady Children’s Hospital, San Diego, California; Thomas A. Coffelt, MD, and Ben Bauer, MD, Riley Hospital for Children, Indianapolis, Indiana; Nada Harik, MD, Arkansas Children’s Hospital, Little Rock; Mythili Srinivasan, MD, St Louis Children’s Hospital, St Louis, Missouri; Ilana Waynik, MD, Connecticut Children’s Medical Center, Hartford; Brett Anderson, MD, MBA, Children’s Hospital of New York–Presbyterian, New York City, New York; Marcos Mestre, MD, Miami Children’s Hospital, Miami, Florida; Adam Berkwitt, MD, Yale–New Haven Children’s Hospital, New Haven, Connecticut; John Kinnison, MD, and Tiffany Shea Osburn, MD, Children’s Hospital Central California, Madera; Bhanumathy Kumar, MD, Children’s Hospital of Michigan, Detroit; David Kotzbauer, MD, Children’s Healthcare of Atlanta, Atlanta, Georgia; George Hescock, MD, Children’s Hospital New Orleans, New Orleans, Louisiana; Sheilah Snyder, MD, Children’s Hospital and Medical Center, Omaha, Nebraska; Edward Chu, MD, Children’s Hospital Oakland, Oakland, California; Sri Narayanan, MD, Children’s Hospital of Alabama, Birmingham; Bahman Panbehi, MD, Children’s Hospital of Orange County, Orange, California; Nader Shaikh, MD, MPH, Children’s Hospital of Pittsburgh, Pittsburgh, Pennsylvania; Rainer Gedeit, MD, Children’s Hospital of Wisconsin, Milwaukee; Marc Mazade, MD, Cook Children’s Medical Center, Fort Worth, Texas; and Kristen Sheets, MD, and Joni Oberlin, MD, East Tennessee Children’s Hospital, Knoxville.
Additional Contributions: Kathryn Conaboy, BA, and Darlene Barkman, MA, members of the Children’s Hospital of Philadelphia Family Advisory Council, guided us in the selection of patient-centered outcomes for the study. They received no financial compensation for this contribution.
et al. Clinical practice guidelines by the Infectious Diseases Society of America for the treatment of methicillin-resistant Staphylococcus aureus
infections in adults and children: executive summary. Clin Infect Dis
. 2011;52(3):285-292.PubMedGoogle ScholarCrossref
et al. Practice guidelines for outpatient parenteral antimicrobial therapy. Clin Infect Dis
. 2004;38(12):1651-1672.PubMedGoogle ScholarCrossref
MJ; Finnish Study Group. Simplified treatment of acute staphylococcal osteomyelitis of childhood. Pediatrics
. 1997;99(6):846-850.PubMedGoogle ScholarCrossref
et al. Acute bacterial osteoarticular infections: eight-year analysis of C-reactive protein for oral step-down therapy. Pediatrics
. 2012;130(4):e821-e828. doi:10.1542/peds.2012-0220. PubMedGoogle ScholarCrossref
CE. Prospective evaluation of a shortened regimen of treatment for acute osteomyelitis and septic arthritis in children. J Pediatr Orthop
. 2009;29(5):518-525.PubMedGoogle ScholarCrossref
R. Prolonged intravenous therapy versus early transition to oral antimicrobial therapy for acute osteomyelitis in children. Pediatrics
. 2009;123(2):636-642.PubMedGoogle ScholarCrossref
et al; Pediatric Research in Inpatient Settings (PRIS) Network. Prioritization of comparative effectiveness research topics in hospital pediatrics. Arch Pediatr Adolesc Med
. 2012;166(12):1155-1164.PubMedGoogle ScholarCrossref
FA. Pediatric deaths attributable to complex chronic conditions: a population-based study of Washington State, 1980-1997. Pediatrics
. 2000;106(1, pt 2):205-209.PubMedGoogle Scholar
DV. Bootstrap Methods and Their Application. New York, NY: Cambridge University Press; 1997.
CE. Separating between- and within-cluster covariate effects by using conditional and partitioning methods. J R Stat Soc B
. 2006;68:859-872. doi:10.1111/j.1467-9868.2006.00570.x. Google ScholarCrossref
JP. Interval estimation for treatment effects using propensity score matching. Stat Med
. 2006;25(13):2230-2256.PubMedGoogle ScholarCrossref
JM. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press; 2002.
BI. Analysis of Health Surveys. New York, NY: Wiley; 1999.