Impact of Clinical Guidelines on Hospital Utilization in Children With Croup Free

We conducted a cross-sectional study of encounters among children aged ≥3 months to ≤8 years presenting to an ED with a diagnosis of croup (International Classification of Disease, 10th Revision [ICD-10] code J050) between January 1, 2019 and June 30, 2022. This period corresponds to the time when ≥50% of hospitals were using their most recent guideline revision.

We excluded direct admissions, transfers, duplicate encounters (ie, 2 encounters with overlapping admission), and encounters with unclassifiable disposition, missing demographics, or complex chronic condition(s), as identified using Feudtner’s classification system.¹¹  We also excluded encounters with concurrent diagnosis of bronchiolitis, asthma, pneumonia, or croup mimicking conditions (eg, bacterial tracheitis) using ICD-10 codes described in a previous study.¹² 

This study followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for cross-sectional studies and was deemed exempt by our hospital’s Institutional Review Board.

Administrative data were obtained from the Pediatric Health Information System (PHIS), which is managed by the Children’s Hospital Association (Lenexa, KS) and includes data from the largest US children’s hospitals. Data quality and reliability are assured through a joint effort between the Children’s Hospital Association and participating hospitals. Participating hospitals provide encounter-level data, including demographics, diagnoses, procedures, and resource utilization measures.

Between January and February 2022, we contacted each PHIS hospital and requested the most recent croup guideline from their database coordinator(s) or clinical staff. Guidelines were categorized as present, absent, or unknown for participating institutions. We defined RE threshold as the minimum number of RE doses that a patient should require before admission according to the guideline. For example, a RE threshold of 2 indicates the recommendation for admission would require at least 2 doses of RE before considering admission. Other descriptors of guidelines included, when available, last revision month and year, original creation month and year, age inclusion criteria, dexamethasone recommendation, specific RE indication, ICU admission criteria, and ED discharge criteria. Guideline criteria were independently reviewed by 2 physician study authors, with a third physician study author independently adjudicating any discordances.

The primary outcome measures were hospital admission rate and costs. Secondary outcomes consisted of resource utilization measures, including ICU admission, neck or chest radiograph utilization, RE and dexamethasone administration, and 7-day related revisits. Hospital admission was defined as discharge from an observation or inpatient unit as opposed to discharge from the ED. Hospital costs were measured as standardized unit costs, which are calculated by PHIS based on the median cost of services rendered for each hospital and represent a proxy for true resource use that is comparable across hospitals.¹³  Hospital costs were also adjusted for inflation using the Consumer Price Index.¹⁴  Because of extreme outliers in hospital costs, we also Windsorized inflation adjusted costs to the 0.5 and 99.5 percentile. Seven-day related revisit was defined as a return to the ED or admission for a condition with same All Patients Refined Diagnosis Related Groups within 7 days of discharge from the initial encounter. ICU admission, neck radiograph, chest radiograph, RE, and dexamethasone administration were categorized as binary outcomes (ie, yes or no) based on presence of charges for these interventions.

Encounter-level demographic characteristics and healthcare utilization outcomes were summarized by presence of a croup guideline and RE threshold (1, 2, or 3 doses). Utilization outcomes by presence of a croup guideline were compared using a 2 sample probability test and t test for dichotomous and continuous outcomes, respectively, and are reported as percentage point differences with 95% confidence intervals (CIs).

Mixed effects logistic and linear regression were used to model dichotomous and continuous outcomes, respectively, with the main explanatory variable of interest being either presence of croup guideline or specific RE threshold. For both the primary and secondary analysis, random intercepts were included to account for within hospital correlation. All models were adjusted for the following covariates: age, sex, race, primary payer, previous croup encounters, and discharge year. Results for our regression models are reported as adjusted odds ratios (aOR) with 95% CIs. We used intraclass correlation coefficients to summarize the proportion of variation in our outcomes explained by random effects.

All analyses were conducted using Stata version 16 (StatCorp. 2019. Stat Statistical Software: Release 16. College Station, TX: Stat Corp LLC) and P values < .05 were considered statistically significant.

A total of 48 hospitals were initially contacted; we excluded 9 for lack of response and 1 because of reporting all encounters as transfers. Of 38 hospitals in our final analysis, 20 had croup guidelines and 18 did not. Of the 20 hospitals with guidelines, the guidelines recommended RE thresholds of 1 (3 of 20 hospitals), 2 (14 of 20 hospitals), and 3 (3 of 20 hospitals) doses.

A total of 374 450 croup encounters were identified, of which 121 284 were included in analysis; 54 102 (44.6%) encounters occurred in hospitals with croup guidelines and 67 182 (55.4%) in hospitals without guidelines (Fig 1). For all encounters, the mean age was 2.1 years, 64.8% were male, and 49.7% identified as non-Hispanic, white race (Table 1).

The overall admission rate was 5.7% (range 0.6% to 18.5%) (Fig 2). The unadjusted admission rate was 2.0 percentage points lower (95% CI −2.3 to −1.7, Table 2) for hospitals with guidelines (4.6%) compared with hospitals without guidelines (6.6%). However, after adjusting for confounders, admission rate was not significantly different for hospitals with guidelines compared with hospitals without guidelines (aOR 0.97, 95% CI 0.58 to 1.64; Table 2). Mean costs were $675 overall and significantly higher for hospitals with guidelines ($704) compared with hospitals without guidelines ($651) (mean difference $53, 95% CI 43 to 63). After adjusting for confounders, mean costs were not significantly different for hospitals with guidelines compared with hospitals without guidelines (mean difference $58, 95% CI −124 to 241). There was no significant difference in secondary outcome measures among hospitals with guidelines compared with hospitals without guidelines after adjusting for covariates.

Among encounters presenting to hospitals with guidelines, admission rates were lower for hospitals with guidelines recommending a minimum of 1 RE (2.3%) compared with hospitals with guidelines recommending a minimum of 2 RE (5.1%) or 3 RE (5.1%) (Supplemental Table 4), however differences were not statistically significant (Table 3). Similarly, mean costs were not significantly different for hospitals with guidelines recommending a minimum of 2 RE (mean difference $26, 95% CI −265 to 316) or 3 RE (mean difference −$67, 95% CI −439 to 305) compared with hospitals with guidelines recommending a minimum of 1 RE (Table 3). Of secondary outcome measures, related revisits within 7 days were significantly lower among hospitals with guidelines recommending a minimum of 2 RE (aOR 0.72, 95% CI 0.55 to 0.96) or 3 RE (aOR 0.71, 95% CI 0.50 to 1.00) compared with hospitals with guidelines recommending a minimum of 1 RE (Table 3).

In this large cross-sectional study across US children’s hospitals, we found admission rates and costs did not differ significantly between hospitals with and without croup guidelines. Similarly, there were not significant differences in secondary outcomes, including ICU admission, radiographs, RE and dexamethasone administration, and 7-day related revisits.

Although the average admission rate of 5.7% for our entire population is within an expected range estimated by previous literature,^1,4,6,9  we noted substantial variation in croup admission rates across individual institutions even when guidelines were present. Mean and median hospital admission rates were 6.0% (interquartile range 3.6–10.2) and 7.0% (SD 4.4), respectively, varying between 0.6% to 18.5% overall and even more substantially for patients receiving at least 1 dose of RE. There are several possible explanations for these findings. Croup is a diagnosis with known marked variation in management and use of NRIRs.^4,5  Although clinical practice guidelines may have standardized aspects of care within institutions, many guidelines not only exist independently of one another, but extrapolate from studies designed for outpatient settings.^4,7  A 2017 study of free-standing children’s hospitals also in the PHIS database, for example, found “up to fivefold difference in NRIR utilization attributable to hospital-level practice variability in inpatient croup care,”⁴  and that variation might extend to admission practices. Additionally, seasonality and geography impact many respiratory conditions such as croup, which is partly reflected in regional differences observed in our own results here and literature elsewhere.⁵  Furthermore, there are other indications for hospital admission not captured in our study, such as dehydration requiring intravenous fluids.

Although our croup admission rates appeared to vary more widely than expected, we uncovered similarities to a prior large multisite PHIS study, which suggested high variability in admission rates for croup.⁴  For example, replicating data from a 2017 study, the lowest and highest 10th percentiles of admission rates found were <1.7% and >14.0%, respectively.⁴  Thus, other literature may have under reported variation in admission trends, or admission rates may be related more fundamentally to limitations on how admissions are defined in study data.

Another potential explanation for admission variation may be the clarity of guideline recommendations themselves, which could impact how effective they are in actual practice. Besides admission rates, we noted differences related to higher dexamethasone use, lower use of radiographs, higher 7-day related revisits, and costs for hospitals with guidelines, although these did not remain significant following statistical adjustment. Collectively, these findings suggest that existing guidelines may not be effectively driving clinical decision making in these hospitals.

The majority of guidelines in our study had an RE threshold of 2, reflecting traditional practices, and only 3 hospitals recommended an RE threshold of 3, despite its growing literature support.^7,8,10  Although we noted differences that could be related to higher perceived croup severity for patients requiring more RE doses, such as use of neck radiographs, these did not remain significant after regression, except for 7-day related revisits. This finding is consistent with literature elsewhere that interventions for patients discharged after 2 RE doses are often limited.^4,7,8,15  However, the low number of hospital guidelines with an RE threshold of 3 in our study precludes our ability to draw meaningful conclusions about impacts on patient outcomes. It may be that guidelines could benefit from de-emphasizing the absolute number of RE doses recommended before admission and instead place more weight on clinical factors.

Overall, as the use of clinical practice guidelines has been supported for a variety of conditions by many studies,^16–18  we expected guidelines for croup to similarly impact clinical outcomes more than observed. The limitations of the PHIS database may hinder the interpretation of our results, however our study could lay the groundwork for a warranted multicenter investigation in the future. Furthermore, rather than demonstrating a limited role for guidelines for croup, guidelines could benefit from emphasizing clinical factors rather than number of RE doses in the decision to admit. Our findings lead us to hypothesize whether existing variability in guideline format approaches should be considered for other conditions, such as gastroenteritis or anaphylaxis, that require similar observation periods for disposition decision-making.

There are several limitations to our study. First, our dataset relied on ICD-10 coding for case ascertainment of croup, and we did not have access to medical records to determine coding accuracy. Second, we were unable to determine whether children received dexamethasone or RE before ED evaluation (ie, in an outside ED, general clinic, or by emergency medical services). Notably, a study conducted in 2019 found that 20% of children received RE before ED evaluation.⁸  Thus, our sample may have underestimated the total number of RE treatments a child may have received and subsequent management. Third, our study period overlapped with the onset of the coronavirus disease 2019 (COVID-19) pandemic, which may have influenced clinical practices. Indeed, co-occurring COVID-19 infection has been associated with increased admission rates and illness severity.^19,20  However, we included discharge year as a covariate in our regression model to account for changes or trends in admission rates over time. Fourth, as PHIS data are premised on billing codification, we defined an admission by inpatient or observation charge, however this may not correlate accurately to true disposition. For example, EDs may have observation units within their own departments that are not billed separately, but nonetheless function as dedicated spaces for patients effectively “admitted” for further management when unable to be discharged. Other hospital systems may send patients to separate observation units, which effectively might be considered boarding at other institutions. We could not locate studies using PHIS data for comparable conditions with observation periods defining admissions differently and in a manner replicable for our study.^21,22  Length of stay is not available in PHIS beyond number of days, limiting ability to use hours in ED as a proxy for understanding observation patterns. Additionally, because of dataset limitations, we were unable to assess diurnal effects on admission, which appears to influence clinical practices.²³  Fifth, hospital effects appear to account for a significant proportion of variation in admission. We considered comparing outcomes pre and postguideline implementation within individual hospitals to bolster our study design; however, this became practically infeasible because even the most recently updated versions of guidelines were difficult to obtain and many could not be confirmed. Finally, our study did not ascertain clinician adherence to guideline recommendations, which given their wide variation in readability, including our own attempts at categorizing, may deviate even further in actual practice. Although the COVID-19 pandemic, guideline adherence, and readability may have influenced outcomes related to croup incidence, admissions, and resource utilization, our data represents a broad national sample of hospitals to mitigate these possible confounding effects.

Croup clinical practice guideline management recommendations and hospital admissions vary substantially across US children’s hospitals. Admission rates, costs, and other metrics did not differ significantly between hospitals with and without croup guidelines, however dataset limitations may hinder interpretation of results. Future studies could incorporate additional strategies, such as multicenter investigations or the use of other markers of resource use such as length of stay to overcome variation in local billing and coding practices.