Variation in the Management of Hospitalized Children With Orbital Cellulitis Over 10 Years

We conducted a retrospective multicenter cohort study of children admitted to hospital with severe orbital infections between January 1, 2009, and December 31, 2018. The study protocol has been previously published.¹⁰  Medical records of children admitted to community (n = 3) and children’s (n = 7) hospitals in the Canadian Pediatric Inpatient Research Network (https://www.pirncanada.com) were reviewed. Ethics approval was obtained from all participating sites. The study was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology checklist.¹⁵ 

Data sources included hospital charts, health record databases, and diagnostic imaging databases. Trained research assistants at each site reviewed eligible patients for study inclusion and abstracted relevant variables using standardized case report forms. Data were stored in the Research Electronic Data Capture online database managed at 1 children’s hospital.¹⁶ 

Children aged 2 months to 18 years hospitalized at each participating site were identified using the International Classification of Diseases (ICD), 10th Revision, Canada, H05.0 diagnostic code (acute inflammation of the eye), which includes periorbital cellulitis, orbital cellulitis, orbital abscess, and subperiosteal abscess. Repeat admissions beyond 30 days of the index admission were considered a unique visit. Because the focus was on previously healthy patients, children were excluded if they had the following diagnoses: orbital tumor or pseudotumor; herpes simplex or herpes zoster infection; previous craniofacial or ocular surgery; craniofacial anatomic abnormality; cellulitis related to trauma, laceration, or recent surgery; underlying acquired or congenital lesion; and immunodeficiency or immunocompromised status.

Patient demographics included age (years), sex, chronic disease status, history of orbital infections, transfer from another hospital, and antibiotics before admission. Chronic disease was self-reported in the medical chart and defined by involved systems (eg, neurologic). Clinical signs at initial presentation included maximum emergency department (ED) temperature recorded, and the presence or absence of proptosis, painful extra ocular movements, ophthalmoplegia, abnormal vision, and eye swollen shut.

Diagnostic tests extracted included complete blood cell (CBC) count, white blood cell count (x10⁹/L), C-reactive protein (CRP) level, erythrocyte sedimentation rate (ESR), blood culture, and ocular discharge culture. Cross-sectional imaging included completion of computed tomography (CT) and magnetic resonance imaging (MRI) scan. If performed, the initial CT scan was used to classify patients according to the Chandler Criteria, a radiographic score for evaluating the anatomic structures and associated complications in orbital cellulitis. A Chandler score of I refers to periorbital cellulitis; II, orbital cellulitis; III, subperiosteal abscess; IV, intraorbital abscess; and V, cavernous sinus thrombosis.¹⁷ 

Data on treatments administered in hospital were also collected, including administration and type of empiric intravenous antibiotics, and adjunctive agents including intranasal decongestants (eg, xylometazoline hydrochloride), intranasal corticosteroids (eg, beclomethasone nasal spray), intranasal saline spray, and systemic corticosteroids (eg, dexamethasone). Antibiotics were classified similar to Krueger et al (C.K., E.N., S. M., unpublished data), into 7 groups:

• Second-generation cephalosporin alone (ie, cefuroxime)

• Third-generation cephalosporin alone (ie, ceftriaxone)

• Third-generation cephalosporin and antianaerobic agent (ie, metronidazole)

• Third-generation cephalosporin and antistaphylococcal agent (ie, cloxacillin, vancomycin)

• Third-generation cephalosporin with both antianaerobic and antistaphylococcal agents

• Antistaphylococcal agent alone (ie, cloxacillin, vancomycin)

• Other

Data on surgical interventions and subspecialty consults were extracted. Clinical outcomes included length of stay (LOS) (hours), revisits to the ED within 30 days, and readmissions at revisits.

Baseline demographic characteristics and study outcomes were summarized descriptively. Continuous variables were summarized as medians with interquartile ranges (IQRs) or as means with SD. Categorical variables were summarized with frequencies or percentages. Where appropriate, 95% confidence intervals were provided. Hospitals were coded with a number from 1 to 10 to preserve confidentiality.

χ² tests were performed to determine whether observed values exceeded that expected by chance for diagnostic testing, imaging, adjunctive agents, antibiotic selection, surgical intervention, and clinical outcomes. Our main analysis reported differences across all sites and among hospital type (community versus children’s hospitals). We then conducted a planned secondary analysis to specifically evaluate for differences among children’s hospitals only. In these analyses, we included only hospitals with a minimum of 25 cases to enable adequate comparisons (ie, sites 1–6, 8–10).¹²  Site 7 was excluded because it had 4 cases. The Mann-Whitney U test was used to analyze differences in LOS among hospital type. The Kruskal-Wallis test was performed to analyze differences in LOS among only the 6 children’s hospitals.

To describe hospital-level variation among the children’s hospitals compared with the median across hospitals, rates of diagnostic tests, adjunctive agents, antibiotics, and surgical intervention were illustrated as a heat map, highlighting hospitals that were >20% below median, 10% to 20% below median, median (within 10% above/below), 10% to 20% above median, and >20% above median. Given the small number of hospitals (n = 6) and lack of standardized score for severity of illness, no risk adjustment was completed for these comparisons.

To examine the association between clinical management and LOS, we conducted a median regression analysis with the following exposures: CT scan, MRI scan, CBC, electrolytes, CRP, ESR, blood culture, ocular discharge culture, and subspeciality consult (ophthalmology, otolaryngology, neurosurgery, infectious diseases). Covariates included age (years), sex (male/female), continuous white blood cell count (x10⁹/L), transfer from a community to a children’s hospital (yes/no), hospital type (community versus children’s), and Chandler score. We also conducted an additional analysis where the exposure was hospital type (children’s or community) and the outcome was LOS with fixed effects for hospital. Patients were removed if they were missing exposure data or had duplicate encounters. Missing covariate data were handled by multiple imputation using the fully conditional specification method.

Trends over time for diagnostic tests, interventions, and outcomes were described at the patient level. Simple linear regression with 95% confidence intervals of the slope was used to evaluate for statistically significant trends in management over time.

All analyses were performed using GraphPad Prism (version 8.0.0 for MacOS, GraphPad Software, San Diego, California). A 2-tailed P value <.05 was considered statistically significant.

We identified 1579 patients across 7 children’s and 3 community hospitals (Supplemental Fig 4). Of 1579 patients, 271 (17.2%) were admitted to a community hospital (Table 1). Median age was 5.4 years (IQR 2.4–9.9) with a male predominance (59.5%), and half (50.3%) received antibiotics before the ED visit. There were significant differences (P < .001) in Chandler Criteria I, II, III, and IV among patients at children’s versus community hospitals (Table 2). Median Canadian Triage and Acuity Scale score in the ED was 3 and did not differ significantly among hospital type (P = .99). Characteristics of transferred patients can be found in Supplemental Table 6.

Variation in management was initially evaluated across all hospital types (Supplemental Table 3). The most ordered diagnostic tests, CBC, electrolytes, and blood culture, had minimal differences across hospital sites in observed values (CBC, median 97.3%, IQR 96.1%–100%; electrolytes, median 81.6%, IQR 73.6%–87.3%; blood culture, median 79.3%, IQR 74.3%–83.4%) and rates of testing (CBC, P = .91; electrolytes, P = .10; blood culture, P = .49).

Among the children’s hospitals with sufficient patient volume for inclusion (sites 1–6), significant variation was observed as illustrated in Fig 1, Fig 2, and Supplemental Fig 5. There were significant differences in rates of ordering CRP tests (P < .001), ESR tests (P < .001), and ocular discharge cultures (P < .001), with comparatively few differences in rates of CBC, electrolytes, and blood cultures. Children’s hospitals also differed in rates of diagnostic imaging, with significant variation in both CT scans (P = .004) and MRI scans (P = .003). There were also important differences in subspecialty consulting rates for otolaryngology (P < .001), ophthalmology (P < .001), and infectious diseases (P < .001), but not neurosurgery (P = .44). There was a larger observed range for CT scans (median 68.5%, IQR 62.7%–78.9%) compared with MRI scans (median 8.7%, IQR 7.0%–10.2%).

We observed significant differences in usage of intranasal decongestants (P < .001), intranasal corticosteroids (P = .003), intranasal saline spray (P < .001), and systemic corticosteroids (P < .001). Broad spectrum empiric antibiotic regimens with dedicated anaerobic and/or staphylococcal coverage were the most frequently administered: third-generation cephalosporin with antianaerobic and staphylococcal agents (group E; 25.9%), third-generation cephalosporin with antistaphylococcal agent (group D; 24.3%), and third-generation cephalosporin with antianaerobic agent (group C; 17.6%). There were significant differences in use of nearly all antibiotic groups among children’s hospitals (Fig 2). Surgical intervention also varied across the children’s hospitals (P < .001), with a threefold difference between the lowest (8%) and highest (24%) rates. The children’s hospitals demonstrated significant differences in return visits to ED within 30 days (P < .001), readmission to hospital at revisit (P = .003), complications (P < .001), and median LOS (P < .001).

When categorizing sites by hospital type, significant variation remained (Table 2). Community hospitals ordered fewer electrolyte tests, CRP tests, ESR tests, CT scans, and MRI scans compared with children’s hospitals. There was also between 5- to ninefold differences in usage of adjunctive agents and statistically significant variation in empiric antibiotic group usage among hospital types. Community hospitals had a significantly lower median LOS of 56.8 hours (IQR 40.1–81.8) compared with 88.5 hours (IQR 53.7–144.0) at children’s hospitals (P < .001). Community hospitals also had a lower rate of complications compared with children’s hospitals (1.8% vs 6.0%; P = .007). There were no significant differences in return visits to ED within 30 days and readmission to hospital at revisit among hospital type.

The results of the median regression analyses are depicted in Supplemental Table 4. Children’s hospital patients had significantly longer LOS compared with patients at community hospitals (P = .001). When the analysis was restricted to children’s hospitals only, patients receiving blood cultures, CT scans, MRI scans, electrolyte tests, CRP tests, and ESR tests had a significantly longer LOS compared with patients who did not. CBC tests and ocular discharge cultures showed no significant association with LOS. All subspeciality consults were associated with a longer LOS (P < .001).

Changes over time in management are outlined in Fig 3 and Supplemental Table 5. Between 2009 and 2018, the rates of CBC and electrolyte testing remained unchanged. However, CRP tests increased significantly from 28.8% to 73.5% (P < .001) whereas ESR tests decreased from 31.5% to 14.1% (P < .001). Rates of CT scans remained similar over time, whereas MRI scans doubled from 4.1% to 8.8% (P = 0.02).

There were also significant changes over time in usage of adjunctive agents, with increases from 37.8% to 57.6% in intranasal corticosteroids (P = 0.003) and 33.6% to 52.9% in intranasal saline sprays (P = .02). Use of intranasal decongestants and systemic corticosteroids were similar. The use of third-generation cephalosporins with both antianaerobic and antistaphylococcal agents increased 21.7%, with a parallel 6.8% decrease in narrow-spectrum therapy (second-generation cephalosporin alone). Surgical intervention rates remained between 10% and 15% between 2009 and 2018. Median LOS and other clinical outcomes were similar over time (Supplemental Table 5).

In this large, multicenter, national cohort of hospitalized children with orbital infections, we observed substantial variation in management and outcomes across multiple children’s and community hospitals. There was significant variation in several diagnostic tests and treatments among hospitals, with notable changes in management over time, particularly rising usage of tests and adjunctive agents. Increased diagnostic test usage was also associated with a longer LOS. Despite the wide variation across sites and over time, clinical outcomes, including LOS and surgical intervention, were unchanged.

Although some diagnostic test usage, such as CBC and electrolytes, had limited variation across sites and, over time, there was significant variation in usage of CRP and ESR. From 2009 to 2018, there was a threefold increase in CRP usage, with a concomitant fall in ESR. In 2018, nearly three-quarters of the study population had a CRP test completed. The rise in CRP usage may reflect its growing recognition as a useful marker of inflammation in acute infections, such as orbital cellulitis,^18,19  with ESR reserved for chronic inflammatory states. In children with severe orbital infections, elevated initial CRP has been identified as a predictor of surgical intervention, of discriminating patients with orbital cellulitis from those with periorbital cellulitis,^20,21  and is correlated with longer length of hospital stay.²²  However, the clinical utility and value of CRP has only been recognized recently, and likely does not explain the rise in usage and the variation among children’s hospitals in its use. Over time, the increased availability of CRP, coupled with CRP’s higher sensitivity, may explain its increase.

Published cross-sectional imaging rates for hospitalized children with severe orbital infections varies from 20% to 75%^4,9,23  which was similar to the wide variation observed in our study. Although the median CT scan rate of 68.5% (IQR 62.7%–78.9%) in the 6 children’s hospitals is comparable to a rate of 74.7% (IQR 66.7%–81.0%) in a study of 42 US children’s hospitals,⁹  there was wide variation. One hospital reported a CT scan rate of 47.9%, whereas another reported 84.6%, suggesting differences in thresholds to scan or in diagnostic certainty of the treating clinician. For example, in pediatric head trauma, head CT scan rates are higher in community hospitals compared with children’s hospitals,²⁴  but can be reduced with quality improvement initiatives.²⁵  Nonpediatric care providers may be less comfortable assessing children, particularly for difficult signs (eg, extraocular movement), and be more likely to order imaging. Interestingly, we observed lower rates of CT scans in community hospitals, which likely reflects differences in severity of presentations rather than experience. Lastly, hospitals with high CT usage were associated with high rates of surgical intervention without any change in overall clinical outcomes.⁹  Given the importance of minimizing unnecessary radiation exposure, our findings highlight the importance of developing standardized imaging guidelines for children who present with suspected orbital infections to reduce potentially unnecessary usage of CT scans. We also observed a rise in MRI scans over time, which was also observed by a single-center study,⁴  which may reflect additional utility in complex cases or increased availability.

We observed significant variation and changes over time in use of intranasal steroids, intranasal decongestants, and intranasal saline.^26,27  These agents are used empirically to treat the underlying etiology of sinusitis, which is present in most cases, reduce mucosal edema, and improve sinus drainage. In certain studies, all hospitalized patients are given intranasal steroids,²⁸  whereas in others, use of these agents vary.^4,29  Evidence supporting the use of these agents is conflicting,^30–32  and no clinical practice guidelines exist.²⁶  In our study, intranasal saline spray and intranasal corticosteroids increased significantly over the 10-year period, despite a limited evidence base. The significant differences in hospital otolaryngology consult rates may have impacted the decision to initiate intranasal therapies. The increased usage of these agents over time despite the lack of effectiveness and safety data highlights the need for further research on intranasal medications.

We observed significant differences in empiric antibiotics used by hospital type (children’s versus community) and across children’s hospitals. Hospitals varied in use of third-generation cephalosporins, antianaerobic agents, and antistaphylococcal agents, consistent with a study reporting >200 different variations of antibiotics used in US children hospitalized with orbital cellulitis.⁹  Interestingly, cefuroxime monotherapy did not vary significantly, which could reflect overall consensus of its limited use compared with third-generation cephalosporins.⁴  Ceftriaxone has been found to be superior to cefuroxime in the treatment of bacterial meningitis, a serious complication of orbital cellulitis.³³  Furthermore, over the 10-year study period, there was a twofold increase in patients receiving empiric antibiotic regimens containing both staphylococcal and anaerobic coverage, with a decline in narrow-spectrum therapies (eg, cefuroxime). A recent analysis using data from this cohort suggests that broader coverage is not associated with improved clinical outcomes (C.K., E.N., S. M., unpublished data) supporting observations noted in single-center studies.⁴  Ultimately, the variations identified highlight the need for consensus-based guidelines for the selection of empiric IV antibiotics.

The variation observed among sites is partially because of differences in hospital type. Community hospitals ordered significantly fewer diagnostic tests, prescribed fewer adjunctive agents, and had a shorter median LOS. These observations are likely because of differences in the patient populations, with more severe and complex infections managed at children’s hospitals. Significant variation was also observed even when comparing management across children’s hospitals. Despite significant changes over time, clinical outcomes remained similar. Interestingly, these findings contrast with US studies, which report increased surgical intervention rates over time.³⁴  Potential explanations include differences in types of hospitals and patient characteristics, or a lower threshold for surgical intervention. Interestingly, rates of surgical intervention varied nearly threefold among Canadian children’s hospitals. The consistent clinical outcomes over time questions whether the increases in testing, imaging, and adjunctive agents may reflect overuse.

Our study has several relevant limitations. First, the recorded data set is limited to patients admitted before 2019, limiting our ability to determine whether trends persisted over time. Second, the retrospective study design limits the study to data fields which are available in the patient chart. Third, we used ICD codes to identify eligible patients and may have missed individuals that were incorrectly coded. Additionally, we were unable to adjust for hospital-level factors across sites given the small number of hospitals, or for severity of illness given the lack of a standardized tool available for all hospitalized patients, unlike that used by others.⁹  Lastly, site 7 had a lower number of cases identified compared with the other sites. However, to ensure data quality and assurance, we double-checked the ICD, 10th Revision, codes locally and sought additional clinical input at the site to ensure the numbers were accurate.

Our study identified notable variation in the management of hospitalized children with orbital cellulitis across hospitals, hospital type, and children’s hospitals. We also revealed increases in test usage and intervention rates over time despite minimal change in patient outcomes. Our findings highlight an opportunity for international site-level collaborations, with an aim to develop much-needed evidence-based clinical practice guidelines for severe orbital infections in children.