To describe variation in the care of children hospitalized with orbital cellulitis and to determine associations with length of stay (LOS), emergency department (ED) revisits, and hospital readmissions.
By using the Pediatric Health Information System, we performed a multicenter, retrospective study of children aged 2 months to 18 years with a primary International Classification of Diseases, Ninth Revision, Clinical Modification discharge diagnosis code for orbital cellulitis from 2007 to 2014. We assessed hospital-level variation in the use of diagnostic tests, corticosteroids, and antibiotics individually and in aggregate for association with outcomes (LOS, ED revisits, readmissions) after risk-adjusting for important clinical and demographic factors.
A total of 1828 children met inclusion criteria. Complete blood cell counts (median [interquartile range]: 81.8% [66.7–89.6]), C-reactive protein levels (57.1% [22.2–84.0]), blood cultures (57.9% [48.9–63.6]), and computed tomography imaging (74.7% [66.7–81.0]) were the most frequently performed diagnostic tests, with significant variation observed across hospitals (all P < .001). Corticosteroids were used in 29.2% of children (interquartile range: 18.4–37.5). There was significant variation in antibiotic exposure across hospitals (P < .001). Increased total diagnostic test usage was associated with increased LOS (P = .044), but not with 30-day ED revisits (P = .176) or readmissions (P = .403).
Children hospitalized with orbital cellulitis experience wide variation in clinical management. Increased hospital-level usage is associated with increased LOS. Our findings highlight a critical need to identify treatment strategies that optimize resource use and outcomes for children hospitalized with orbital cellulitis.
Orbital cellulitis infections are serious and can lead to life-altering conditions, including vision loss, meningitis, brain abscess, and cavernous sinus thrombosis.1–3 These infections account for an estimated 2525 hospital admissions in children per year within the United States, with estimated aggregate annual charges of $49.3 million.4 Although several hospitals have implemented local clinical practice guidelines and diagnostic algorithms, no national guidelines or clinical trials have established optimal diagnostic or treatment regimens in the United States.5–7 Without consensus guidelines, there is likely great variation in clinical practice within and across hospitals. Investigations of diagnostic testing and management of other disease processes, including pneumonia and diabetic ketoacidosis, have demonstrated that increased variation and resource use is associated with increased rates of hospitalization, higher costs, and increased length of stay (LOS).8–14 However, similar studies focused on variation in management and clinical outcomes of orbital infections are lacking.
Identifying variation in care and its relationship to patient outcomes is an important step in the national effort to improve health care delivery and patient outcomes.15 The aims of this study are to describe hospital-level variation in diagnostic testing, corticosteroid use, and antibiotic selection in a large, national cohort of children with orbital cellulitis and to describe the association of variation in resource use with LOS, emergency department (ED) revisits, and hospital readmissions.
Study Design and Data Source
This is a multicenter, retrospective study of children hospitalized with a principal International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) discharge diagnosis code of 376.01 (orbital cellulitis). Data were derived from the Pediatric Health Information System (PHIS), an administrative and billing database of 46 tertiary care pediatric hospitals affiliated with the Children’s Hospital Association (Lenexa, KS). Patient data have been deidentified within PHIS; however, encryption of patient identifiers allows for tracking of individual patients across visits. The current study included data from 42 hospitals, with 4 hospitals excluded for billing data quality concerns. This study was approved by the local institutional review board.
Children aged 2 months to 18 years who were admitted to a PHIS-participating hospital from January 1, 2007, through December 31, 2014, with a principal ICD-9-CM discharge diagnosis code of 376.01 were eligible for inclusion. If patients had multiple hospitalizations within a 30-day period, only the index hospitalization was included.
To identify children with orbital cellulitis who were otherwise healthy, we excluded patients with congenital malformations, prematurity, low birth weight, underlying malnutrition, or complex chronic conditions (Supplemental Table 2).16 Children with competing ophthalmologic diagnoses were excluded. Children with underlying diagnoses that would increase the likelihood of corticosteroid administration (eg, asthma, adrenal insufficiency) were excluded to allow for examination of corticosteroid use. Children with secondary diagnoses of intracranial abscess and trauma were excluded because these children would be unlikely to undergo management for orbital cellulitis alone. Children who received antifungal or antiviral therapy, and those who did not receive systemic antibiotics within the first 2 days of hospitalization, were excluded for the possibility of a nonbacterial infection. Because of a high proportion of preseptal infections identified during chart review with our initial inclusion and exclusion criteria, we subsequently excluded children discharged within the first 2 days because of their increased likelihood for preseptal infection.
An internal validation study was performed through manual chart review at 2 PHIS hospitals to assess the accuracy of our case identification strategy. Medical records with a principal ICD-9-CM code of 376.01 were reviewed by board-certified pediatricians and classified as periorbital cellulitis (ie, preseptal infection) versus orbital cellulitis with or without abscess (ie, post-septal infection). Diagnostic category assessment was confirmed by a second reviewer, with discrepancies adjudicated by a board-certified pediatric infectious diseases physician. Of the 468 records reviewed, 301 met inclusion criteria. Of included records, our identification strategy was associated with a positive predictive value for identifying orbital infection of any type (pre- or postseptal infection) of 100.0% and a positive predictive value for orbital cellulitis with or without abscess of 89.7%. Among children who underwent surgical intervention, the positive predictive value for orbital cellulitis with or without abscess was 98.4%.
All records were assessed for the presence of a billing code for selected diagnostic tests performed within the first 2 days of hospitalization. Laboratory measures included complete blood cell (CBC) count, C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), chemistries, blood cultures, cerebrospinal fluid (CSF) cultures, fungal wound cultures, and bacterial wound cultures. Imaging studies included computed tomography (CT), MRI, and other imaging. Other imaging was defined as radiographs or ultrasound images of the orbits, sinuses, or head.
Corticosteroid and Antibiotic Exposure
Corticosteroid use was defined as oral or parenteral administration of at least 1 of the following during the first 2 days of hospitalization: dexamethasone, methylprednisolone, prednisolone, or prednisone. Empirical antibiotic selection was defined as antibiotics administered during the first 2 days of hospitalization and was divided into 2 broad categories: only oral or any parenteral. Use within the first 2 days of hospitalization was chosen to most closely correlate with empirical antibiotic choice and to minimize the influence of microbiological test results on antibiotic choice, a factor that could not be captured in the PHIS database. Antibiotic selection was defined by using a hierarchical algorithm (Supplemental Fig 5, Supplemental Table 3). Within our study, children who received oral preparations alone were categorized as only oral. Children who received parenteral antibiotics were then categorized on the basis of the antibiotic’s spectrum of activity. Children who received vancomycin, daptomycin, or linezolid were defined as vancomycin alone or in combination. Children who received clindamycin but did not receive a β-lactam/β-lactamase inhibitor antibiotic were defined as clindamycin alone or in combination. Children who received clindamycin and a β-lactam/β-lactamase inhibitor were defined as clindamycin/β-lactam combination. Children who did not receive clindamycin but received a β-lactam/β-lactamase inhibitor antibiotic were defined as β-lactam alone or in combination. Finally, children who did not receive clindamycin or a β-lactam/β-lactamase inhibitor antibiotic typically prescribed for the management of orbital or sinus infection were defined as other antibiotic. Categorization was reviewed and confirmed by the 2 board-certified pediatric infectious diseases physicians in the study group.
Resource Use and Outcome Measures
Resource use included hospital-level rates of diagnostic testing, corticosteroid use, and antibiotic selection. Outcome measures included LOS, 30-day ED revisit rates, and 30-day readmission rates. Only ED revisits or readmissions secondary to complications of orbital cellulitis or its management were considered (Supplemental Table 4).
Demographics and Clinical Characteristics
Collected demographics and clinical characteristics included age, sex, race and/or ethnicity, primary payer admission season, region of the United States, surgical intervention, peripherally inserted central catheter (PICC) placement, ICU transfer, and case mix index (CMI). PICC placement was defined by the presence of a corresponding procedural code for catheter placement. Surgical intervention was defined by the presence of procedural codes for ophthalmic or sinonasal surgical procedures. ICU transfer and CMI were chosen to serve as surrogate markers for the severity of illness (SOI). ICU transfer was defined as any child requiring transfer to the ICU after admission. CMI in PHIS is a relative weight assigned to each discharge on the basis of the All-Patient Refined Diagnostic Group (APR-DRG) (3M, St Paul, MN) assignment and APR-DRG SOI, which ranges from 1 (minor) to 4 (extreme). The weights are derived by Truven Health Analytics (Ann Arbor, MI) from its nationally-representative pediatric database as the ratio of the average charge for discharges within a specific APR-DRG and SOI combination to the average charge for all discharges in the database. For simplicity of reporting and interpretation, we split the weights at the median into 2 groups: minor and major.
Continuous variables were summarized with medians and interquartile ranges (IQRs), whereas categorical variables were summarized with frequencies and percentages. We calculated hospital-level summary statistics to assess variation in diagnostic test use, corticosteroid use, and antibiotic selection across hospitals, and made comparisons across hospitals by using the χ2 test or Kruskal-Wallis test. Hospital-level diagnostic test usage rates and outcomes were risk-adjusted for age and continuous CMI by using appropriate distributions for generalized linear mixed effects models with a random intercept for each hospital. After adjustment, hospitals were assigned risk-adjusted diagnostic test usage scores for individual diagnostic tests (ie, a hospital-level usage score for an individual diagnostic test) on the basis of their absolute percentage above and below the median rate. For each diagnostic test, hospitals received 2 points if their risk-adjusted rate was >10% over the median, 1 point if it was 5% to 10% over the median, 0 points if it was within 5% of the median, −1 point if it was 5% to 10% under the median, and −2 points if it was >10% under the median. We then assigned a total diagnostic test usage score to each hospital by summing the risk-adjusted diagnostic test scores for individual tests (ie, we generated an aggregate score for use of all diagnostic tests at an individual hospital). We correlated the risk-adjusted total diagnostic test usage score with the risk-adjusted outcomes by using Pearson’s correlation coefficient. We similarly correlated risk-adjusted total diagnostic test usage scores with risk-adjusted outcomes in a substudy analysis of children with orbital cellulitis who underwent surgical intervention. All statistical analyses were performed by using SAS version 9.4 (SAS Institute, Cary, NC), and P values < .05 were considered statistically significant.
A total of 1828 children across 42 hospitals met the inclusion criteria (Fig 1). Of these, 70.8% were <10 years of age, 66.0% were boys, and 2.1% required transfer to the ICU (Table 1). The median CMI for the study population was 0.95 (IQR: 0.77–2.49). Among children included in our cohort, 23.9% underwent surgical intervention. The median percentage of children across hospitals that underwent surgical intervention was 50.0% (IQR: 23.4%–76.6%).
|Characteristic .||n (%) .|
|2 mo–1 y||301 (16.5)|
|2–4 y||386 (21.1)|
|5–9 y||608 (33.3)|
|10–14 y||443 (24.2)|
|15–18 y||90 (4.9)|
|Race and/or ethnicity|
|Non-Hispanic white||933 (51.0)|
|Non-Hispanic African American||398 (21.8)|
|Asian American||38 (2.1)|
|Characteristic .||n (%) .|
|2 mo–1 y||301 (16.5)|
|2–4 y||386 (21.1)|
|5–9 y||608 (33.3)|
|10–14 y||443 (24.2)|
|15–18 y||90 (4.9)|
|Race and/or ethnicity|
|Non-Hispanic white||933 (51.0)|
|Non-Hispanic African American||398 (21.8)|
|Asian American||38 (2.1)|
Surgical intervention was defined on the basis of the presence or absence of procedural codes for ophthalmic or sinonasal surgical procedures.
PICC placement was defined on the basis of the presence or absence of a corresponding procedural code for catheter placement.
ICU transfer was defined as any child requiring transfer to the ICU after admission.
CMI is a relative weight assigned to each discharge on the basis of the ARP-DRG assignment and ARP-DRG SOI. For simplicity of reporting and interpretation, we split the weights at the median into 2 groups.
Variation in Resource Use and Association With LOS, ED Revisits, and Readmission
We observed substantial variation in diagnostic test use across hospitals (Figs 2 and 3). CBC counts, CRP tests, blood cultures, and CT scans were the most frequently performed diagnostic tests. CSF cultures, MRIs, and other imaging were obtained infrequently.
Variation in the treatment of orbital cellulitis with systemic corticosteroids and antibiotics was also observed across hospitals. Use of adjunctive corticosteroids occurred in a subset of pediatric patients (29.2% [IQR: 18.4–37.5], P < .001), with use observed within 41 of the 42 hospitals included in the study. Antibiotic selection varied widely across hospitals with more than 200 unique combinations of antibiotics prescribed in the first 2 days of hospitalization before categorization. The most frequently prescribed antibiotic regimens included vancomycin alone or in combination (43.1% [IQR: 29.2–62.9], P < .001) and clindamycin/β-lactam combinations (30.9% [IQR: 14.3–54.2], P < .001). Twelve of the 1828 (0.7%) children within the cohort had exposure to antibiotics included within the parenteral other antibiotic category, whereas 1 (0.05%) child received oral antibiotics alone. Variation in antibiotic selection was also observed within hospitals, as demonstrated by the distribution of antibiotic groups within a majority of hospitals (Fig 3). We observed similar variation in diagnostic test use, corticosteroid use, and antibiotic selection in a subanalysis of children with orbital cellulitis who underwent surgical intervention (Supplemental Figs 6 and 7).
After risk adjustment for age and continuous CMI, there was significant variation across hospitals in LOS (4.1 days [IQR: 3.8–4.2], P < .001), but not for 30-day readmission rates (1.6% [IQR: 0.0–3.5], P = .082), or 30-day ED revisit rates (2.3% [IQR: 0.0–4.7], P = .053). An increased total diagnostic test usage score at the hospital-level was significantly associated with an increased LOS (P = .044) (Fig 4). For every 5-point increase in the score, the risk-adjusted LOS increased by 0.19 days (95% confidence interval: 0.01–0.37). No statistically significant association was found between the total diagnostic test usage score and the risk of 30-day ED revisits (P = .176) or 30-day readmissions (P = .403). We observed similar variation in LOS across hospitals in a subanalysis of children with orbital cellulitis who underwent surgical intervention (Supplemental Fig 8).
In this multicenter, retrospective study, we observed substantial variation across hospitals in diagnostic test use, corticosteroid use, and empirical antibiotic selection in a large, national cohort of children with orbital cellulitis. Extensive variation in empirical antibiotic selection was observed not only across hospitals but also within individual hospitals. We found that increased diagnostic test use was associated with increased LOS but was not associated with ED revisits or hospital readmissions.
Significant variation was observed for the use of nearly every diagnostic test in our cohort. The variation we observed may reflect clinical uncertainty of the utility of some diagnostic tests in predicting outcomes for patients with orbital cellulitis. For example, although the authors of some studies have suggested that elevated white blood cell counts are associated with the presence of a drainable abscess, authors of other studies have shown that the degree of elevation does not reliably predict the development of a complicated course or the need for surgical intervention.17,18 Additionally, expert opinion varies regarding the appropriate use of blood cultures in the evaluation of orbital cellulitis, challenging the formation of a consensus guideline.3,5,7,19 The authors of previous studies have reported that the rates of blood culture positivity among children with orbital cellulitis range from 2.27% to 7%.5,19–22 However, studies have revealed that rates of blood culture contamination among pediatric patients range from 1% to 4%,23–25 making it difficult to develop a universal recommendation for obtaining a blood culture in the setting of low likelihood of positivity in patients with orbital cellulitis.
Within our study, we observed significant variation in the use of corticosteroids and in the choice of empirical antibiotic for children hospitalized with orbital cellulitis. The authors of previous studies have suggested that using adjunctive corticosteroids in the management of orbital cellulitis may be beneficial; however, the generalizability of these findings to all pediatric patients is limited by small sample sizes derived from single instiutions.26–28 Within our study, corticosteroids were used in 29.2% (IQR: 18.4–37.5) of children, with use observed in 41 hospitals, suggesting that corticosteroids were used for a minority of patients at many of the hospitals. The variation in corticosteroid use may reflect the lack of strong evidence for clinical benefit as reported in the literature, as well as reflecting physician concern for the adverse effect profile of corticosteroids, particularly for host immune suppression and the increased risk for infectious complications. The variation in empirical antibiotic selection that we observed is consistent with observations from previous studies of orbital cellulitis.1,18,20,29,30 The wide variation in empirical antibiotic selection that we observed both within and across hospitals may similarly reflect a lack of a clear consensus recommendation as to optimal empirical antibiotic choice. Regardless, the use of such a diverse range of antibiotic combinations highlights the uncertainty clinicians face when choosing optimal treatment regimens for orbital cellulitis.
Higher diagnostic test use at the hospital level was associated with increased LOS, a finding that occurs in other serious infections in children.8–14 Not only are prolonged hospitalizations more costly, but they also increase the risk of acquiring nosocomial infections, place an increased psychosocial burden upon families, and reduce the quality of life for the hospitalized child.31–34 Although one might infer that increased test use might provide more comprehensive management, and therefore decrease rates of revisits, we did not observe any significant reduction in revisits among hospitals with high diagnostic testing usage versus hospitals with low diagnostic testing usage, even after controlling for age and SOI. Understanding the significant variation in pediatric orbital cellulitis hospitalizations may prove influential in the development of clinical guidelines aimed at improving the quality of care delivered in this subset of infections, because guideline development has proven to be beneficial within other disease processes.35
This study has several important limitations. First, the ICD-9-CM discharge diagnosis code 376.01 encompasses a range of orbital infections, including periorbital cellulitis, orbital cellulitis, orbital abscess, and subperiosteal orbital abscess. Consequently, some of the variation in care we observed may reflect differences in admission practices and in how clinicians manage pre- versus postseptal infections. Although there are limitations in using the ICD-9-CM diagnosis code to identify orbital cellulitis, this is a previously published strategy.36 Additionally, we attempted to mitigate misclassification bias by identifying those most likely to have postseptal infection through our application of exclusion criteria and through examination of a cohort of children who underwent surgical intervention. Although our case definition includes both pre- and postseptal orbital infections, our internal test of validity revealed that our identification strategy was associated with a positive predictive value of 89.7% for postseptal disease in our general population and with a positive predictive value of 98.4% among those who underwent surgical intervention. As the validation occurred at 2 hospitals, the results may have differed (higher or lower positive predictive value) with the inclusion of additional hospitals. Our substudy analysis of children who underwent surgical intervention (ie, children with high likelihood for postseptal disease), revealed similar results on Pearson’s correlation compared with the entire study population, further supporting our case definition’s validity. Within our study, LOS was used as an exclusion criterion as well as an outcome measure. Although this methodology has the potential to bias toward detecting a difference, the results of the analyses were unchanged regardless of whether children hospitalized for <2 days were included or excluded. The PHIS database contains administrative data only, thus limiting our ability to evaluate the association of patient presentation with clinical decision-making. Particularly, we had limited ability to assess orbital cellulitis severity or illness severity, microbiologic test results, or local patterns of antimicrobial resistance, factors that might have influenced LOS and decision-making, including diagnostic test use and antimicrobial prescription practices. We attempted to mitigate these effects by assessing surrogate markers for illness severity, such as CMI, and by focusing on empirical antibiotic selection. Finally, the differences in LOS observed between hospitals were modest, and additional factors, including confounding from unmeasured variables, may have contributed to the association between diagnostic testing and LOS that we observed.
Our study provides a comprehensive overview of the wide variation in clinical management that children hospitalized with orbital cellulitis experience. With our findings, we highlight an opportunity for antimicrobial stewardship for orbital cellulitis and a definitive need for future prospective investigations aimed at identifying optimal diagnosis and management strategies. Future studies examining patient-level factors in association with outcomes may more clearly identify opportunities for targeted management strategies and help guide the development of diagnostic and treatment guidelines for orbital cellulitis.
Dr Markham conceptualized and designed the study, analyzed and interpreted the data, drafted the initial manuscript, and coordinated all edits of the manuscript. Dr Hall was involved in the study design, supervised the data analysis and interpretation, and reviewed and revised the manuscript as submitted. Drs Bettenhausen, Myers, and Puls assisted with study design, participated in the interpretation of data, and reviewed and revised the manuscript. Dr McCulloh supervised the conceptualization and design of the study, participated in the interpretation of data, and reviewed and revised the manuscript; and all authors approved the final version of the manuscript as submitted.
FUNDING: No external funding.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.