Data Source

We used the Pediatric Health Information System (PHIS), an administrative database that contains inpatient data from 40 free-standing children’s hospitals affiliated with the Child Health Corporation of America (CHCA, Overland Park, Kansas). Contributing hospitals are located in 17 of the 20 major metropolitan areas in the United States and account for 70% of all freestanding children’s hospitals in the US (Data from The National Association of Children’s Hospitals and Related Institutions (Alexandria, VA). The database includes detailed information on demographics, diagnoses, procedures, medications, and repeat hospitalizations. Included in the medication files are data on the type and route of administration of all antimicrobials administered during hospitalization. Oversight of PHIS data quality and accuracy is a joint effort between CHCA, Thomson Healthcare (the data manager), and participating hospitals. Data are de-identified at the time of data submission and subjected to 175 reliability and validity checks. Data are accepted into the database when classified errors occur in fewer than 2% of a hospital’s quarterly data.

Assembly of Study Cohort

Our study included children ages 2 months to 17 years with discharge dates between January 1, 2000 and June 30, 2005. Children were included in the study cohort if their index hospitalization was assigned an International Classification of Diseases (ICD-9-CM) diagnosis code for acute osteomyelitis or unspecified osteomyelitis (730.01–730.09 and 730.2–730.29) in any of the 21 diagnosis fields. Children with a hospitalization for chronic osteomyelitis in the 6 months prior to the index admission were excluded. In addition, children were excluded if they were discharged during time periods when CHCA deemed a hospital’s data invalid or missing.

To define a cohort of children with uncomplicated, acute osteomyelitis, we excluded children with specific comorbid conditions documented on the index or prior admissions and those who were hospitalized for 10 or more days. Exclusionary comorbid conditions included those whose presence suggests complicated or difficult to treat osteomylelitis; congenital and acquired immunodeficiencies, sickle cell disease, trauma, osteomyelitis associated with immobilization or pressure ulcers (e.g., spina bifida, quadriplegia, paraplegia, mechanical ventilation, post-operative infections) and osteomyelitis of the head, face and orbits. We also excluded children with conditions that would predispose to inadequate absorption of oral medications (e.g., malabsorption).

In addition, we excluded children with an ICD-9-CM code for any of the following conditions during any admission prior to study entry that might have increased the risk of subsequent complicated osteomyelitis: cellulitis, pyogenic arthritis, sacroilitis, synovitis, myositis, chronic sinusitis, arthropathy, congenital or acquired diseases of bone, fasciitis, postoperative wounds, and placement of orthopedic devices or prosthesis. Finally, we excluded children with less than 6 months of observation time after the initial admission for osteomyelitis, to allow adequate time for observation of study outcomes.

Exposure classification

Children were classified into one of two possible treatment groups at the time of discharge from the hospital-- prolonged intravenous antimicrobial therapy or early transition to oral antimicrobial therapy. The prolonged intravenous therapy group was defined by the presence of a procedure code of 38.93 (venous catheterization, not elsewhere classified), which represents the placement of a central venous catheter. Children without procedure code 38.93 were assumed to have been discharged on oral therapy and comprised the early transition to oral therapy group. We assumed that transition to oral therapy occurred at the time of discharge, which may represent the most conservative estimate of transition time. The assignment of children to the prolonged intravenous or early transition to oral therapy groups was validated by CHCA as part of a data validation project using a 10% random sample of the study cohort from 19 of the 29 hospitals that agreed to participate in the validation study. The charts of children included in the validation sample were reviewed to confirm placement of a central venous catheter during the index hospitalization or early transition to oral antimicrobial therapy as part of the discharge plan.

Outcome measures

The primary outcome was treatment failure, defined as re-hospitalization within 6 months with assigned diagnosis or procedure codes consistent with (1) acute osteomyelitis as the sole diagnosis (730.0–09), (2) chronic osteomyelitis (730.1X), (3) a potential complication of acute osteomyelitis (e.g, myositis, arthritis, etc.), or (4) a surgical procedure related to the musculoskeletal system. Secondary outcomes included re-hospitalization within 6 months for (1) any reason; (2) catheter related complication; and (3) adverse drug reactions associated with antibiotics, C. difficile infection or agranulocytosis, a not uncommon effect of beta-lactam antibiotics. To determine the outcomes, two authors (TZ and RK) who were blinded to treatment group assignment reviewed the ICD-9-CM diagnosis and procedure codes for all children who were re-hospitalized during the study period.

Covariates

We extracted data from the PHIS database on the following potential confounders and effect modifiers: age, sex, race, surgical procedure, anatomic location of the infection, and the presence of an ICD9-CM code for Staphylococcus aureus (S. aureus) infection or methicillin-resistant S. aureus (MRSA). We also collected information on severity of initial illness using the PHIS case mix index (CMI), a widely used risk adjustment measure based upon relative weights derived from Thomson Healthcare’s national pediatric charges per case data and 3M’s All-Patient-Refined Diagnosis-Related Group (APR-DRG) classification system. The relative weights are computed as the ratio of the average charges per patient in an APR-DRG/Severity of Illness group to the average charges for all other children.

Statistical Analysis

Summary statistics were constructed using frequencies and proportions for categorical data elements and means and medians for continuous variables. The chi-square test was used for unadjusted comparisons between children who received prolonged intravenous therapy and children who received oral therapy. Random effects models and likelihood ratio tests were used to determine the significance of inter-hospital variation in use of oral therapy as well as in treatment failure rates. A propensity score model was developed to balance patient-level confounders that may have determined the selected treatment strategy. Propensity score analysis attempts to identify children who are similar except for their treatment or exposure status. The scores represent the probability that a patient will receive a treatment strategy based on his or her observed covariates.^{9, 10} We calculated propensity scores using multivariable logistic regression with early transition to oral therapy as the outcome of interest. Scores were then grouped into quintiles for later use a covariate in the analysis of mode of administration and treatment failure. In the response model (a logistic regression with early re-hospitalization as the outcome and quintile of propensity score as a covariate), we introduced singly each covariate from the propensity score to identify any residual confounding from that covariate.¹¹ All confidence intervals were adjusted for the clustering of children within hospital using robust variance estimates.¹²

To determine the effect of exposure misclassification on the observed association between mode of antimicrobial administration and treatment failure, we compared results for the subset of hospitals that participated in the validation study and had no misclassification of exposure with results for the hospitals that did not participate in the validation study or were found to have some degree of misclassification of the exposure identified in the validation study.

We performed additional analyses to determine the within and among hospital effects. Confounding by hospital can occur when both the exposure of interest and outcome are clustered by hospital. To address this potential problem, we decomposed the within- and among-hospital components of the effect of mode of antibiotic administration.^{13, 14} The within-hospital effect measures the association of early transition to oral therapy and outcome once a child has selected a hospital and been admitted. The among-hospital effect measures the impact on outcome of transferring a given child from a hospital with lower to higher oral therapy use. ¹⁵ All analyses were conducted with SAS version 9.1 (Cary, NC) and Stata statistical software version 8.0 (College Station, TX).

Treatment failure

The overall treatment failure rate among the 1969 children was 4.7% (95% confidence interval=3.8% to 5.7%) (Table 2). The treatment failure rate varied across hospitals from as low as 1.8% to as high as 12.5% for hospitals with at least 10 children with acute osteomyelitis, but this variation was not statistically significant (likelihood ratio chi square <0.001, p=0.50). The treatment failure rate was 5% (54 of 1021) in the prolonged intravenous therapy group and 4% (38 of 948) in the oral therapy group. There was no significant association between treatment failure and the mode of antimicrobial therapy adjusted for propensity score quintiles and clustering of observations within the 29 hospitals in the sample (Odds Ratio=0.77, 95% confidence interval=0.49 to 1.22) (Table 2). The median time to treatment failure was similar in both groups, 16.5 days (interquartile range, 6–35 days) for the prolonged intravenous therapy group compared to 14 days (interquartile range, 3–45 days) for the oral therapy group (p=0.65). Further, we did not detect any residual confounding after re-inclusion of all covariates from the propensity score in the final multivariate analysis.

Table 2

Treatment Outcomes of Acute Osteomyelitis.^*

Primary Outcome	Intravenous Therapy (1021) N (%)	Oral Therapy (948) N (%)	Propensity-score adjusted odds ratio (95% confidence interval) for those children treated with early transition to oral therapy
Treatment Failure within 6 months of diagnosis	54 (5)	38 (4)	0.77 (0.49, 1.22)
Chronic osteomyelitis	13 (1.3)	8 (0.8)	0.84 (0.33, 2.13)
Musculoskeletal surgery	18 (1.8)	15 (1.6)	0.80 (0.38, 1.70)
Complication of osteomyelitis**	11 (1.1)	6 (0.6)	0.75 (0.27, 2.07)
Acute osteomyelitis as sole readmission diagnosis	12 (1.2)	9 (0.9)	0.72 (0.25, 2.08)
Secondary Outcomes
Any re-hospitalization within 6 months of diagnosis	102 (10)	56 (5.9)	0.6 (0.38, 0.96)
Catheter-associated complication	35 (3)	0 (0)	---
Adverse effect of antimicrobials^	15 (1.5)	4 (0.4)	0.39 (0.14, 1.1)

Open in a separate window

^*Analysis used the complete cohort.

^**Includes synovitis, pyogenic arthritis, sacroilitis, disorders of bone and cartiliage NOS, and disc disorder

^{^} adverse drug reactions associated with antibiotics, C. difficile infection or agranulocytosis

Validation study

Nineteen of the 29 hospitals agreed to participate in the validation study. Thirteen hospitals found no misclassification of the exposure and 6 hospitals found that some children who were treated with prolonged intravenous therapy were misclassified and assigned to the oral therapy group. The rate of misclassification for these hospitals ranged from 11 to 50%. None of the hospitals reported misclassification of children assigned to the intravenous therapy group. Among the hospitals reporting no misclassification, the odds ratio for treatment failure of oral therapy compared with intravenous therapy was 0.74 (95% confidence interval, 0.27 to 2.02). Similar results were observed in the cohort of 16 hospitals that did not participate in the validation study or reported misclassification of the exposure (odds ratio, 0.73; 95% confidence interval: 0.47 to 1.13).

Within and among hospital effects

The effect of mode of antimicrobial administration was decomposed into within and among hospital effects. The within hospital effect favoring oral therapy was 0.73 (95% confidence interval: 0.39, 1.35). There was also no among-hospital association of oral therapy and outcome (odds ratio, 0.98; 95% confidence interval: 0.93, 1.04).

Financial support. Dr. Theoklis E. Zaoutis was supported by grant K23 AI0629753-01 from the National Institutes of Health. Dr. A. Russell Localio was supported in part by an Agency for Healthcare Research and Quality (AHRQ) Centers for Education and Research on Therapeutics cooperative agreement (grant #HS10399). Dr. Ron Keren was supported by grant K23 HD043179 from the National Institute of Child Health and Human Development, Bethesda, MD.