Association Between Corticosteroids and Outcomes in Children Hospitalized With Orbital Cellulitis

We conducted a retrospective cohort study using data obtained from the Pediatric Health Information System (PHIS), an administrative database containing combined hospitalization data from 49 free standing US children’s hospitals developed by the Children’s Hospital Association (Lenexa, KS). The database contains detailed data on demographics, diagnosis codes, service locations, procedures, and charges. All patient-level data are deidentified within the PHIS database, but study subjects are linkable across encounters. The lead author’s research ethics board approved this study. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.¹¹ 

Children between the ages of 2 months and 18 years hospitalized at a PHIS-participating hospital from January 1, 2007, to December 31, 2019, with a primary discharge diagnosis of orbital cellulitis were eligible for inclusion. Eligible patients were identified on the basis of International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) or International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnostic codes. We used a single ICD-9-CM code, 376.01, which has been used extensively in previous studies.^10,12–14  Several refinements were made in ICD-10-CM, and we included diagnostic codes for orbital cellulitis (H05.011, H05.012, H05.013, H05.019, H05.021, H05.022, H05.023, H05.029) and periorbital cellulitis (L03.213). Given the strong clinical overlap in hospitalized children, periorbital and orbital cellulitis are considered together and, for simplicity, will be referred to as orbital cellulitis. If there were multiple admissions within a 30-day period, only the first hospitalization was included.

To focus on otherwise healthy children with orbital cellulitis, we excluded children with chronic complex conditions,¹⁵  congenital malformations, history of prematurity or low birth weight, intracranial infections, neoplasms or personal history of malignancy, and nutritional deficiencies (Supplemental Table 3). Children with other related or competing ophthalmologic conditions (eg retinal detachment), eye-related infections (eg herpes simplex virus keratitis), and secondary diagnoses of trauma were also excluded because management would be unlikely related to orbital cellulitis alone. We excluded children with conditions that would increase the likelihood of concomitant systemic corticosteroid use (eg asthma, adrenal insufficiency) to avoid the possibility that corticosteroid treatment was for another condition. Children who had surgery on the first day of hospitalization were excluded (day 0), because these patients already experienced the outcome of interest and exposure to corticosteroids would not have changed the outcome. We also excluded children who received an antifungal or antiviral therapy administered within the first 2 days of hospitalization given the possibility of a nonbacterial or more complicated infection. Lastly, we excluded children who did not receive a systemic antibiotic within the first 2 days of hospitalization or who only received topical or oral antibiotics.

The primary exposure was systemic corticosteroid administration as part of the initial medical management of hospitalized children with orbital cellulitis. Corticosteroids were identified by using billing data and included either enteral or parental formulations of dexamethasone, prednisone, prednisolone, or methylprednisolone, which are the most commonly used agents. Initial medical management was defined as corticosteroid administration on day 0 (the initial day of hospitalization). The noncorticosteroid group was defined as those who did not receive corticosteroid therapy on day 0. The rationale for the exposure definitions was to differentiate the use of corticosteroids as part of initial medical management (eg to avoid need for surgical intervention) from corticosteroids used for perioperative or postoperative care (ie day of or after surgery) or as part of rescue therapy due to failure to respond (ie day 1 or later).

The primary outcome was length of hospital stay, defined as the number of days between date of admission and date of discharge, given previous researchers suggesting shortened recovery time and reduced length of stay with corticosteroids.^5,6,8  Length of hospital stay is an important outcome because it represents the effectiveness of managing the condition and is meaningful to families as a marker of the burden of hospitalization.

The main secondary outcome was the need for surgical intervention, defined as orbital and sinonasal surgery. All procedure codes included in the cohort were inspected, and only those related to orbital and sinonasal surgery were used to define surgical intervention. In Supplemental Table 4 we outline a list of ICD-9-CM and ICD-10-CM codes. Additional secondary outcomes included ICU admission as a marker of severity. All-cause emergency department (ED) revisits and all-cause readmissions at 7, 14, and 30 days were evaluated as surrogate outcomes for treatment failures or worsened clinical outcomes. Health care use outcomes included hospital costs (estimated from charges using annual hospital-specific cost-to-charge ratios) at the index visit and overall episode costs at 7, 14, and 30 days.

Patient demographics included age, sex, race and ethnicity, and payer. Disposition was defined as either home or other, which included home health, skilled facilities, or other. We included season of hospitalization given previous association with seasonal variation in orbital cellulitis hospitalization.¹⁰  Hospitals were assigned to 1 of 4 census regions: Midwest, Northeast, South, and West. Severity of illness was described by using Hospitalization Resource Intensity Scores for Kids (H-RISK), which assigns a relative weight to each discharge using the All-Patient Refined Diagnostic Group (3M) assignment and All-Patient Refined Diagnostic Group severity of illness measure.¹⁶ 

Empirical antibiotics used within the first 2 days of hospitalization were grouped into 5 mutually exclusive categories on the basis of a previously used classification system with modification to exclude oral antibiotics¹⁰ : (1) clindamycin alone or in combination with a non β-lactam antibiotic (eg clindamycin alone or clindamycin and metronidazole); (2) β-lactam alone or in combination (eg ceftriaxone alone or ceftriaxone and metronidazole); (3) clindamycin and β-lactam/β-lactamase inhibitor combination (eg clindamycin and ceftriaxone); (4) vancomycin, daptomycin, or linezolid alone or in combination (eg vancomycin alone or vancomycin and metronidazole); or (5) other antibiotic (eg azithromycin). The other antibiotic category included macrolides, tetracyclines, rifampin, trimethoprim/sulfamethoxazole, quinolones, and β-lactam antibiotics not typically used for orbital or sinus infection. Classification was based on the broadest spectrum of activity of administered antibiotics.

Continuous variables were summarized as median with IQR, and categorical variables were summarized as frequencies with percentages. Groups were compared by using Wilcoxon rank tests or χ² test as appropriate. We used generalized linear mixed effects models with a random intercept for each hospital and inverse probability weighting to adjust for covariates and assess for differences in clinical outcomes. Continuous outcomes were log transformed before modeling with normal distributions, and binary outcomes were modeled with binomial distributions. Results are reported as adjusted odds ratios (aORs) and adjusted rate ratios (aRRs) with 95% confidence intervals (CIs). The propensity to receive corticosteroids for each encounter was derived from a logistic regression by using age, sex, payer, season, and H-RISK. To evaluate for the possibility of residual confounding by indication (ie, severely ill children were more likely to receive broad spectrum antibiotics and more likely to have worse outcomes not adequately accounted for in our primary analysis), we conducted a subgroup analysis restricting the main analysis to only those patients who received broad spectrum antibiotics in group 3 (clindamycin and β-lactam/β-lactamase inhibitor combination) or group 4 (vancomycin, linezolid, daptomycin alone or in combination). We conducted 2 sensitivity analyses evaluating the timing of corticosteroid exposure and surgical intervention. We first broadened corticosteroid exposure to include day 0 or 1 of hospital admission and then further excluded children who had surgery on day 1 to address the possibility of both corticosteroid use and surgical intervention occurring on day 1. Lastly, given the clinical concern regarding differences in periorbital and orbital cellulitis, we conducted sensitivity analyses looking at those with only a diagnostic code for periorbital cellulitis or orbital cellulitis separately. Because of coding practice changes between ICD-9-CM and ICD-10-CM, these sensitivity analyses were only possible in patients with an ICD-10-CM code. All statistical analyses were performed by using SAS v.9.4. (SAS Institute, Inc, Cary, NC), and P values <.05 were considered statistically significant.

We identified 5832 patients who were hospitalized with orbital cellulitis and met study inclusion criteria (Fig 1). Median age was 5.0 (IQR: 2.0–9.0), with 45.9% of children <4 years of age. There was a male predominance (60.8%), and only 1% of patients required admission to the ICU. Initial corticosteroid use across the PHIS study hospitals varied (Fig 2), with a median use of 4.8% (IQR: 1.8–9.4). Of the 49 hospitals, 4 hospitals did not report any use, whereas 12 hospitals reported use in >10% of patients.

There were 330 patients (5.7%) in the day 0 corticosteroid group and 5502 (94.3%) in the noncorticosteroid group (Table 1). There were no differences in sex, race and ethnicity, or season between groups. However, children who received corticosteroid group were more likely to be older, live in the South, have more severe illness, and receive broad spectrum antibiotics (group 3 or 4).

The results of the adjusted analyses are shown in Table 2 (unadjusted in Supplemental Table 5). In the adjusted analysis, early corticosteroid exposure was not significantly associated with length of hospital stay, need for surgical intervention, ICU admission rates, or ED revisits. Children who received corticosteroids did not have a significantly lower odds of readmission to hospital at 7 or 30 days and did not have significantly different episode costs (Table 2).

The results of the subgroup and sensitivity analyses are shown in Fig 3 (Supplemental Table 6). Restricting the main analysis to only those patients who received broad spectrum antibiotics in group 3 or group 4 revealed similar findings to the primary analysis. When corticosteroid exposure was broadened to include day 0 or 1 of hospital admission, corticosteroids were associated with an increased length of hospital stay (aRR: 1.4, 95%: CI: 1.2–1.6), hospital costs (aRR: 1.5, 95% CI: 1.3–1.8), and 30-day episode costs (aRR: 1.6, 95% CI: 1.3–1.8). There was also an eightfold increased risk of surgical intervention (aOR: 8.1, 95% CI: 4–16.4) in children who received corticosteroids on day 0 or 1. To address the possibility of both corticosteroid use and surgical intervention occurring on day 1, we further excluded children who had surgery on day 1. In this revised analysis, the risk of surgical intervention no longer remained significant whereas the increased length of hospital stay, index hospital admission costs and 30-day episode costs persisted. When the analyses were restricted to only those patients with an ICD-10-CM code for periorbital cellulitis or orbital cellulitis separately, there were similar findings to the primary analysis (Supplemental Table 7).

In this large, multisite study of >5800 hospitalized children with orbital cellulitis from 49 US children’s hospitals, we found no evidence of association with early use of systemic corticosteroids and clinically important differences in length of hospital stay, surgical intervention, ED revisit rates, readmission rates, or cost. Our study population focused on previously healthy children, without important comorbidities or potentially competing medical diagnoses, managed on parenteral antibiotic therapy. Our findings suggest that early corticosteroid use may not confer benefit in previously healthy children. In the context of previous work, our findings suggest equipoise and emphasize the need for randomized controlled trials to determine best practice.

In previous studies, researchers suggested that corticosteroids can reduce orbital inflammation and edema, leading to faster recovery time in acute unwell hospitalized children.^5–8  In 2 prospective observational studies of hospitalized children, authors reported a 3-day reduction in length of stay.^5,7  However, in Davies et al,⁶  families were offered corticosteroids once the inflammatory marker C-reactive protein was <4 mg/dL, suggesting patients were already clinically improving. The rate of surgical intervention in the corticosteroid group (54%) was also higher than in the noncorticosteroid group (29%). Similarly, a small randomized controlled trial of 21 patients aged >10 years in a tertiary eye center in India reported a reduction in mean length of stay from 18 to 14 days in participants treated with corticosteroids given at day 4 onwards, after an initial response to intravenous antibiotics, along with less pain, periorbital edema, fever, and chemosis.⁸  Congruent with a recent systematic review, we did not find differences in outcomes in our primary analysis to support routine use of corticosteroids in children with orbital cellulitis.¹⁷  Indeed, researchers in other studies have identified that increased hospital-level resource use, including use of systemic corticosteroids, in hospitalized children with orbital cellulitis was associated with increased length of stay.¹⁸ 

Adjunctive corticosteroids are not universally beneficial in acute bacterial infections. For example, corticosteroids for bacterial meningitis are beneficial only in children with Haemophilus influenzae infection but not Streptococcus pneumoniae and Neisseria meningitidis.¹⁹  Similarly, it is possible that adjunctive corticosteroids may be beneficial in a subset of patients with orbital cellulitis, such as those with a specific bacterial pathogen or with certain demographic factors. Although we adjusted for H-RISK index as a marker of severity and other key demographics, we were unable to account for indication, dosing, timing, or duration of corticosteroid therapy, which could explain the observed findings.

Children who received corticosteroids were more likely to receive broad spectrum antibiotics and had a higher severity of illness. These findings are consistent with previous retrospective studies in which researchers reported children who received corticosteroids had more severe disease, higher inflammatory markers, and higher need for combination antibiotic therapy.¹⁰  However, in the subgroup analysis of only those patients who received broad spectrum antibiotic therapy, which represents the most unwell patients, there was no association with corticosteroid use and improved outcomes. It is possible that there are other differences in these 2 populations, such as other unmeasured covariates, which could be minimized in a prospective study.

Early systemic corticosteroid therapy was only initiated in 5.6% of the study population, which is lower than the 29.2% reported in a large multicenter, retrospective PHIS study of 1818 children hospitalized at US children’s hospitals that applied a similar inclusion criteria.²⁰  Our findings are consistent with studies from the United Kingdom, Portugal, and Canada in which researchers report low rates of corticosteroid use.^20–23  No previous studies differentiated whether the primary indication for corticosteroid use was to reduce orbital cellulitis-related complications or to reduce complications related to surgery. In a survey on prescribing behavior among rhinologists and pediatric otolaryngologist, 45% of respondents reported using systemic corticosteroids in children with orbital complications of acute bacterial rhinosinusitis.²⁴  Furthermore, otolaryngologists or ophthalmologists often initiate corticosteroid therapy pre-, peri-, or postoperatively to reduce swelling and inflammation in children requiring surgery.²⁵  Although our study did not focus on corticosteroid use related to surgery, the findings from the sensitivity analyses of increased length of stay, return ED visits, readmissions, and costs raise uncertainty about the effectiveness of this practice and suggest potential harm.

This study has a number of important limitations. First, administrative data lack important clinical information, including physical examination findings, diagnostic test results, and indication for corticosteroid therapy. In particular, the PHIS database does not include information on clinical outcome measures, such as resolution of symptoms, vision loss, pain, or proptosis. Second, there is a potential for misclassification using ICD codes. ICD-9-CM 376.01 has been used extensively in previous studies, including being validated in a chart review.^10,12,13,26  No researchers using administrative databases have used ICD-10-CM codes, other than in an ongoing multisite retrospective study,²⁷  but we included a broad set of codes to ensure relevant patients were included. We also attempted to reduce misclassification by excluding children with secondary diagnoses, with other potential competing use of corticosteroids, or those who did not receive antibiotics within the first 2 days of hospitalization. Third, given the limitations in administrative data, we were unable to differentiate children with periorbital or orbital cellulitis but conducted a subgroup analysis only looking at those in group 3 or 4 antibiotics, which showed consistent findings. Fourth, we did not have information on indication, dosing, timing, or duration of corticosteroid therapy. We also were unable to obtain information on adverse effects, such as immune suppression or adrenal insufficiency with corticosteroid therapy. Fifth, although we used H-RISK to adjust for severity of illness, unaccounted-for differences in disease severity may have contributed to our findings. The limited number and depth of covariates available in PHIS means that a nontrivial amount of residual confounding may remain. Lastly, the PHIS database only includes children’s hospitals, so our findings may not be generalizable to children hospitalized within other settings. However, given the scant literature on corticosteroid therapy for orbital cellulitis, our study adds meaningful information to the body of literature.

Early use of adjunctive corticosteroids in hospitalized children with orbital cellulitis is not associated with reduced length of stay, decreased need for surgical intervention, or reduced cost. With our study, we provide further evidence of uncertainty and clinical equipoise regarding corticosteroid use and reaffirm the need to conduct prospective, multicenter studies and clinical trials to better evaluate the effectiveness and safety profile of corticosteroids. Urgent evidence-based guidelines are needed to standardize clinical management for children with orbital cellulitis.

We thank Thaksha Thavam for her assistance in creating the figures for the manuscript.