Variation in Emergency Department Use of Racemic Epinephrine and Associated Outcomes for Croup Free

We performed a retrospective cohort study using the Pediatric Health Information System (PHIS), an administrative database that contains data from approximately 50 not-for-profit, tertiary care, pediatric hospitals in the United States that are affiliated with the Children’s Hospital Association (Lenexa, KS). The PHIS data include inpatient, ED, ambulatory surgery and observation encounters that are deidentified, and subject-to-data quality checks.¹¹  The hospitals with complete billing and clinical data for the entire study period were included.

We included children aged 3 months to 10 years who were treated in the ED between January 1, 2016, and December 31, 2020, and had a diagnostic code for croup (International Classification of Diseases, 10th Revision [ICD-10], codes J04 and J05). These ICD-10 codes have been shown to have high sensitivity and positive predictive value for croup.¹² 

We excluded patients with a complex chronic condition, defined using the diagnosis code classification schema reported by Feudtner et al,¹³  given that their medical complexity would alter clinical management. We also excluded patients with diagnoses of congenital anomaly of larynx or trachea, burns, airway foreign bodies, bacterial pneumonia, and retropharyngeal, parapharyngeal, or peritonsillar abscess, because these diagnoses could also involve the airway but would warrant different treatments and likely have a different clinical course than croup.

We extracted patient encounter-level data from the PHIS database, including demographics (age, sex, race, ethnicity, and insurance status), encounter diagnoses, ED medications administered (inhaled RE or systemic steroids), ED arrival hour, patient disposition from the ED (discharge versus hospital admission), and return visits within 3 days. The PHIS database identifies the calendar day a medication was administered but does not distinguish between ED and inpatient charges. For admitted patients arriving before 6 pm, only charges from the date of presentation were included (day 0). For patients arriving at 6 pm or later, charges from the initial and following day were included (day 0 and day 1), because the ED encounter may have bridged multiple calendar days. For patients discharged from the ED, we included all medications and testing for the encounter. This definition has been used previously for PHIS studies in the ED.^14,15  For age, we categorized into the following groups as recommended by the US National Institute of Child Health and Human Development: <1 year, 1 year, 2 to 5 years, and ≥6 years.¹⁶  For encounters without a specified race or insurance type, we classified as “not reported” for purposes of analysis.

The primary outcome was the hospital-specific ED use of RE for patients with croup, measured as the percentage of patients in a given ED receiving at least 1 dose. Secondary outcomes included hospital admission, ICU admission, and 3-day ED revisit.

To determine the variation in RE use among PHIS hospital EDs, we conducted an analysis to differentiate whether RE use was predicted by patient factors or by the hospital in which the care was delivered. We selected a priori the following patient factors present in the PHIS database to evaluate: age group, sex, race, ethnicity, insurance, season, year, arrival hour (categorized as daytime [0800–1559], evening [1600–2359], and overnight [0000–0759]), systemic steroid use in the same patient encounter, and previous encounter for croup within the last 365 days. We chose these variables because age, sex, race, and insurance status have been shown to have associations with severe croup,¹⁷  and croup is known to have diurnal, seasonal, and annual variation.^1,18,19  Steroid use impacts the clinical course of croup,⁴  and previous croup encounter was selected as a potential marker of predisposition to laryngeal edema. Using the selected patient factors, we created 2 separate models. The first was a logistic regression model with only patient factors. We used the results of this model to calculate the expected proportion of patients at each hospital treated with RE on the basis of the hospital’s patient population (case-mix) alone. The second set of models were generalized linear mixed effects models, specifying patient factors as fixed effects and hospital as a random effect. Using these models, we determined the predicted proportions of patients treated with RE on the basis of hospital-specific factors. To visualize how variation in use of RE was related to patient factors and hospital-specific effects, we then plotted the expected proportions on the basis of patient factors alone (case-mix), the predicted proportions on the basis of hospital-specific effects, and the actual observed proportions in the cohort. Finally, we calculated the ratio of hospital predicted values and case-mix predicted values for each hospital and multiplied by the overall proportion of patients in the cohort receiving RE to obtain risk-standardized rates of RE administration for each participating hospital.^2,20  The purpose for generating these risk-standardized rates of RE use was to isolate differences in practice patterns between hospitals, accounting for differences in their respective patient populations.

We then analyzed the relationship between risk-standardized, site-specific ED RE use and the patient-level disposition outcomes of hospital admission, ICU admission, and 3-day ED revisit. We fit generalized linear mixed models for each binary outcome with the site-specific, risk-standardized RE use rate as the primary independent variable and hospital as a random effect to account for clustering. Risk-standardized RE use was a continuous variable in the models, with odds ratios (ORs) calculated on the basis of a unit of 10%. Statistical analyses were conducted using R (version 4.0.2; R Foundation for Statistical Computing, Vienna, Austria) and SAS 9.4 (SAS Institute Inc.).

Over the 5-year study period, we identified 249 246 ED encounters with a discharge diagnosis of croup from 39 PHIS hospitals, of which 231 683 remained after exclusions (Fig 1). The median age of the cohort was 2.0 years (interquartile range [IQR], 1.2–3.6), and 65% were male. A total of 24.7% of patients were treated with RE, 8.6% were admitted to the hospital from the ED, 1.0% required the ICU, and 5.3% had a return visit to the ED or inpatient setting within 3 days. Characteristics of the cohort are shown in Table 1.

The adjusted OR of RE use for each patient characteristic calculated from our mixed effects logistic regression model is shown in Table 1. Male children had greater odds of RE treatment than female children. Children with white race, non-Hispanic ethnicity, and private insurance had greater odds of RE treatment than Black, Hispanic, and publicly insured children, respectively. Children presenting overnight, those with concurrent steroid use, and those with previous croup encounter had increased odds of receiving RE. Use of RE varied on the basis of year and season.

ED use of RE varied by hospital, ranging from 14% to 48%, with a median of 24.5% (IQR, 20.0%–27.8%). Fig 2 illustrates the observed rates of use, as well as rates predicted by patient factors (case-mix) and hospital-specific effects in the model. The rates of RE use predicted by each hospital’s patient characteristics alone showed minimal variation. The variation observed among hospitals persisted after adjusting for available patient characteristics: hospital-specific ED, risk-standardized RE use rates ranged from 14.2% to 44.7%, with a median of 24.5% (IQR, 21.0%–28.4%).

The relationship between risk-standardized, hospital-specific ED use of RE and patient disposition outcomes is shown in Fig 3. We found that increased, site-specific RE use was associated with increased likelihood of inpatient admission. For every 10% increase in ED RE use (standardized to account for patient factors), there was a 39% increase in odds of admission, when adjusting for hospital effects (OR, 1.39; 95% confidence interval [CI], 1.01–1.92). There was a similar relationship between RE use and ICU admission (Fig 3B), though it did not reach statistical significance (OR, 1.40; 95% CI, 0.98–1.98). We did not find any association between hospital-specific ED RE use and 3-day ED revisits (OR, 1.00; 95% CI, 0.92–1.09) (Table 2).

In this multicenter, observational study of children diagnosed in the pediatric ED with croup, we found considerable institutional variation in use of RE after adjusting for case-mix. We found that increased, hospital-specific ED use of RE was associated with increased patient odds of inpatient admission but not ICU admission or 3-day return visits. Patient characteristics were associated with differing odds of an individual patient receiving RE but did not predict rates of site-specific ED use.

Patients treated in hospitals with increased use of RE had increased odds of inpatient admission. This finding may have important implications. Single-center studies have shown that only 20% to 36%^10,21–24  of patients admitted for croup receive any further significant therapy, with rates as low as 14% for patients who were asymptomatic after RE treatment in the ED.¹⁰  Given the questionable benefits of inpatient admission for many patients with croup, it is important to identify and address factors that may lead to increased and potentially unnecessary admissions.

One factor to address may be the considerable variation (14%–48%) in use of RE that exists between hospitals. Our finding that hospital variation exists in the ED use of RE for patients with croup is not surprising given the existing literature evaluating care of children with respiratory diseases. One small study of 2 community EDs demonstrated that physician specialty was associated with significant differences in likelihood of oral versus parenteral steroid, RE, albuterol, and x-ray use in the ED for patients diagnosed with croup.⁹  Inpatient management of croup has been shown to be highly variable between institutions with respect to use of imaging, viral testing, antibiotics, and parenteral steroids.²  Similarly, others have demonstrated hospital variation in the ED management of pediatric respiratory diseases such as pneumonia²⁵  and bronchiolitis.²⁶ 

Although not the driving factor for overall hospital variation in RE use, several patient factors were also associated with RE use. Many of the factors associated with increased odds of RE use included factors known to increase risk for croup, such as male sex, overnight arrival time, and fall season, and thus may be a marker of more severe disease.²⁷  However, we also found differences in receipt of treatment based on race, ethnicity, and socioeconomic factors. Specifically, Black patients had lower odds of receiving the treatment than white patients, Hispanic patients had lower odds than non-Hispanic patients, and publicly insured patients had lower odds than privately insured patients. It is unclear whether these patient factors are markers of severity of illness, or whether these differences are related to the subjectivity behind the decision to treat a croup patient with RE. Further work is needed to examine the relationship between patient factors and important clinical outcomes. Moreover, work is needed to standardize indications for treatment and reduce unintended variation.

Although we did not find a statistically significant association between site-specific RE use and ICU admission, the point estimate of the effect was similar to the point estimate of the association between RE use and hospital admission. It may be difficult to definitively evaluate associations with ICU admission because of the rarity of this outcome. In contrast, there appeared to be no association between site-specific RE use and ED revisits. The increased admission rates at hospitals with greater RE use may bias this association negatively, because hospitalized children cannot revisit the ED while they are hospitalized. However, it is unlikely this is obscuring a true association between RE and increased ED revisits, because the drug is known to have a short duration of action and patients are generally not discharged until the effects of the medication have worn off.

This study has several limitations. As an administrative database, PHIS does not contain clinical information such as physical exam findings or vital signs, so we cannot truly determine the severity of each encounter. Because we cannot determine clinical severity through important markers such as work of breathing, oxygen saturation, or croup scores, it is possible that some hospitals had significantly more severely ill children with croup than others. However, croup is a very common, classic clinical diagnosis, and we excluded diagnoses that would imply a more complex airway history. Although diagnosis codes have been shown to have variable accuracy,²⁸  ICD-10 codes have been shown to have high sensitivity and positive predictive value for croup.¹²  PHIS captures billing charges but we are unable to confirm exact doses of RE patients received. Because PHIS does not distinguish between ED and inpatient charges, it is possible that some patients who received RE only after admission were misclassified as receiving RE in the ED. We do not have access to outpatient prescription data so are unable to determine whether patients had testing or treatment with steroids, RE, or other medications before or after their ED and inpatient hospital encounters. It is likely that a small number of children had revisits at non-PHIS hospitals,²⁹  and although this may mean the true revisit rate is slightly higher than we report, we do not expect this to impact our analysis of variation among hospitals. Our analysis includes encounters from 2020, which bridged the start of the coronavirus disease 2019 pandemic and was an atypical year for respiratory viruses, so may have impacted our results. Finally, our cohort of tertiary care, pediatric hospitals may not be generalizable to general ED settings, because our data and previous PHIS studies^2,23  have shown higher admission rates than other nationwide samples.¹ 

We found considerable hospital variation in the use of RE for croup in the pediatric ED. Increased, site-specific ED use of RE was associated with increased odds of inpatient admission. Further research is needed to determine whether standardization of ED treatment may reduce unintended variation in RE treatment and hospital admission in children with croup.

We thank Dr Benjamin Kerrey and Dr Melinda Mahabee-Gittens for reviewing the manuscript.