Trends in Bronchiolitis ICU Admissions and Ventilation Practices: 2010–2019

This was a retrospective cross-sectional study of PHIS. PHIS is an anonymous, quality-controlled, online administrative data warehouse of >45 US children’s hospitals across >20 states, representing the majority of large US metropolitan centers.^26–28  To ensure that changes seen over time were reflective of practice changes, rather than incorporation of new hospitals, the current study was limited to hospitals that have been providing data since 2010. Patients were eligible if they were discharged between January 1, 2010, and December 31, 2019, from an inpatient hospital stay with an associated primary encounter diagnosis of viral bronchiolitis (International Classification of Diseases, Ninth Revision [ICD-9] 466, or International Classification of Diseases, 10th Revision [ICD-10] J21) and were <2 years of age. There were no exclusion criteria. Because this was an observational study, there was no exposure. The outcomes of interest were IMV, NIV, ICU admission, length of stay, and cost.

Encounter-level data were extracted from PHIS (age, sex, race, ethnicity, diagnostic codes, insurance status, hospital, length of stay, ICU length of stay, complex chronic conditions,²⁹  abstracted charges, and cost). Invasive ventilation and NIV were determined on the basis of diagnostic codes in the methods of Fujiogi et al, supplemental table 3.⁴  Cohort demographics were described by using summary statistics. Cost analyses were abstracted from hospital charges and determined by the cost-to-charge ratio submitted to the Centers for Medicare and Medicaid Services (CMS) annually and adjusted by the CMS wage/price index according to hospital zip code in PHIS.²⁶  Cost-over-time analyses were adjusted for the annual gross domestic product (GDP) or national health care expenditures per capita, as stated in results, and expressed in 2010 dollars.³⁰  To assess changes over time, the cohort was binned into groups on the basis of discharge year, and linear regression was performed, with discharge year as the independent variable. Discharge year, rather than admission year, was chosen to ensure that all patients had complete data at the time of analyses. Mixed-effects multivariable logistic regression was performed, by using a random intercept for hospital region similar to Gupta et al.⁸  Models were developed including age, sex, race, ethnicity, insurance status, complex chronic condition flags, and discharge year. Because age was previously known to display a nonlinear association with IMV in bronchiolitis,⁸  splines regression was performed. Models were simplified by using automated stepwise selection minimizing the Akaike information criterion.³¹  Because assessing changes in practice over time was the primary interest of the study, discharge year was forced into the model even if not significant. No other variables were forced into the models.

All statistical analyses were performed by using RStudio version 1.3.1073 (RStudio, Boston, MA) and R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria) with the following packages: caret, cowplot, dplyr, flexdashboard, forcats, ggplot2, givitiR, jtools, kableExtra, lubridate, precrec, pROC, purr, readr, stringr, tibble, tidyr, tidyverse, and tiff.^32,33  An α value of .05 was set as the threshold for significance. The code used to create the article and supplement, except for single-center analyses, are publicly available at https://github.com/drjonpelly/bronchiolitis_in_phis_supplement.

We conducted 2 sensitivity analyses and 1 nested subanalysis. For sensitivity analyses, we first broadened the cohort to include encounters with any associated admission diagnoses of bronchiolitis, consistent with Fujiogi et al.⁴  Second, we restricted the cohort to only those patients defined within the 3M All Patient Refined Diagnosis Related Group (APR-DRG) (version 32) of “Bronchiolitis and RSV Pneumonia” (group 138, APR-DRG).³⁴  Because bronchiolitis and influenza are increasingly diagnosed by multiplex assays,^35–37  we compared diagnoses of bronchiolitis with influenza (ICD-9 487, ICD-10 J09, ICD-10 J10, or ICD-10 J11). Because there was a rise in coding of “acute respiratory failure” (ICD-9518.81 or ICD-10 J96) with the transition from ICD-9 to ICD-10, we examined the patients with this diagnosis who were also diagnosed with bronchiolitis. For the subanalysis, because PHIS contains neither previously validated severity-of-illness scores^38–40  nor the variables necessary to derive these scores, we analyzed a nested subset of patients from the main analysis admitted to our institution with bronchiolitis. We analyzed the changes in modified electronic Pediatric Logistic Organ Dysfunction score (PELOD) and Pediatric Risk of Mortality score (PRISM) by discharge year for patients admitted to our ICU.⁴¹  The PELOD score was modified to remove points for mechanical ventilation to allow for comparison between groups stratified on level of ventilatory support.

The main analysis included 203 859 admissions with a primary encounter diagnosis of bronchiolitis among 185 658 patients across 38 centers over 10 years. The demographics for the main analysis and all sensitivity analyses are listed in Table 1. Median (interquartile range [IQR]) hospital length of stay was 3 (2–4) days, with 19.3% admitted to the ICU and 8.2% receiving charges for IMV or NIV. The median (IQR) cost per admission was $5224 ($3120–$9365). Overall survival to discharge was 99.9%. The proportion of admissions with complex chronic conditions increased over time, from 14.8% in 2010 to 18.8% in 2019 (Supplemental Information: Demographics).

Bronchiolitis admissions were 10.7% (19 301 of 180 272) of all PHIS hospital admissions in 2010 versus 11.3% (21 747 of 191 812) in 2019 (P = .267 for trend, Fig 1A). They comprised 7.8% (2255 of 28 901) of all PHIS ICU admissions in 2010 versus 13.2% (5334 of 40 498) in 2019 (P < .001 for trend). Overall, 2255 of 19 301 (11.7%) of patients with bronchiolitis were admitted to the ICU in 2010 versus 5334 of 21 747 (24.5%) in 2019 (average increase 1.3% per year, P < .001 for trend, Fig 2A). During the same period, 28 901 of 180 272 (16.0%) of all children <2 years of age in PHIS (any diagnosis) were admitted to the ICU in 2010 versus 40 498 of 191 812 (21.1%) in 2019 (average increase 0.5% per year, P < .001 for trend, Fig 2B).

The proportion of IMV over time did not significantly change (3.3% in 2010 versus 2.8% in 2019, P = .414 for trend, Fig 3A). However, the proportion of NIV increased more than sevenfold over the study period (1.2% in 2010 to 9.5% in 2019, P < .001 for trend, Fig 3A). During this same period, the absolute number of patients admitted to the ICU for IMV numerically decreased from 572 in 2010 to 549 in 2019 (P = .737 for trend, Fig 3B), but the absolute number of patients admitted for NIV or no ventilatory support both significantly increased (169 vs 1599, P < .001 for trend, and 1514 vs 3186, P < .001 for trend, respectively, Fig 3B).

The mixed-effects logistic regression models to predict IMV, NIV, and ICU admission are shown in Table 2. After adjustment for age, race and ethnicity, sex, insurance status, and chronic complex conditions, discharge year was significantly associated with NIV (odds ratio: 1.24; 95% confidence interval [CI]: 1.23–1.24) and ICU admission (odds ratio: 1.09; 95% CI 1.09–1.09) but not IMV (odds ratio: 1.00; 95% CI: 1.00–1.00).

The median (IQR) hospital length of stay did not change between 2010 and 2019: 3 (2–4) days versus 3 (2–4) days, respectively. However, the median (IQR) ICU length of stay decreased from 3 (2–6) days in 2010 versus 2 (1–4) days in 2019 (P < .001 for trend). As seen in Fig 1C, the total GDP-adjusted cost of bronchiolitis diagnoses increased numerically from $145 to $157 million, but this was not statistically significant (average increase $1.1 million per year, P = .276 for trend). The cost of NIV rose from $3.8 million in 2010 to $24.6 million in 2019. The mean GDP-adjusted cost per admission decreased from $7527 to $7219 (average decrease $57 per year, P < .001 for trend, Supplemental Information: Cost). The GDP-adjusted cost per admission decreased over time for all levels of ventilatory support (Supplemental Information: Cost). Additional inflation adjustments are shown in Supplemental Information: Cost.

The complete results of all sensitivity analyses are shown in the Supplemental Information. The proportion of encounters with a primary diagnoses of bronchiolitis (used for the main analysis) did not significantly change over time (10.7% in 2010 versus 11.3% in 2019, P = .267 for trend, Fig 1A). However, there was a significant increase in the proportion of admissions with secondary diagnoses of bronchiolitis (2.9% in 2010 versus 9.7% in 2019, P < .001 for trend). Of the 88 531 patients with a secondary diagnosis of bronchiolitis, 35 140 (39.6%) had a primary diagnosis of acute respiratory failure (639 of 5193 [12.3%] in 2010 versus 11 603 of 18 650 in 2019 [62.2%], P < .001 for trend). Thus, the percentage of patients with all (primary or secondary) diagnosis of bronchiolitis increased from 13.6% in 2010 to 21.0% in 2019 (Fig 1B). Similar trends were noted for influenza (Supplemental Information: Diagnoses).

The proportion of patients admitted to the ICU and treated with IMV and NIV over time displayed similar trends to the main analysis in both the all-diagnosis and the APR-DRG groups (Supplemental Information: Incidence and Supplemental Information: Ventilation). After adjustment for confounders, discharge year remained associated with NIV and ICU admission (Supplemental Information: Model ICU and Supplemental Information: Model NIV) and was positively associated with IMV in the all-diagnosis group (odds ratio 1.02 [95% CI 1.02–1.02]) but not the APR-DRG group (Supplemental Information: Model IMV). In parallel with rising group size, the GDP-adjusted cost for the all-diagnosis group increased from $270 million in 2010 to $466 million in 2019 (average increase $22.6 million per year, P < .001 for trend, Fig 1D), whereas the APR-DRG group was similar to the main analysis ($110 million in 2010 versus $119 million in 2019, Supplemental Inforamation: Cost).

The complete results of the nested single-center subanalysis are shown in Supplemental Information: Single-Center. Mirroring national findings, the proportion of ICU admission in our center rose over the study period, outpacing overall expansion in ICU proportion (rate of change 2.4% per year versus 0.6% per year, P < .001 for both). Our ICU favors HFNC and does not code this as NIV. Thus, the number of patients with a primary diagnosis of bronchiolitis admitted to the ICU for IMV increased by 1.1 patients per year (P = .011 for trend), whereas the number of patients admitted for NIV increased by 0.2 patients per year (P = .852 for trend). However, the number of patients admitted for HFNC therapy increased by 13.5 patients per year (P < .001 for trend). Modified PELOD and PRISM scores were available for patients admitted to our ICU between 2010 and 2018.⁴¹  There was no significant change in the modified PELOD score or the PRISM score for patients admitted to our ICU over the study period.

The present analysis of >200 000 admissions for bronchiolitis over 10 years demonstrates that the ICU admission proportion for bronchiolitis has doubled, as seen in Fig 2. This augments earlier studies by now showing a full decade of steadily climbing ICU burden.⁸  This change outpaced overall ICU admission growth for children <2, which concomitantly rose only 33% over the study period. This is particularly striking when considering that the overall hospital admission rate for bronchiolitis has fallen over the same time period.⁴  Understanding the reasons for increasing ICU admission proportion in this population is clinically important. Patients with bronchiolitis admitted to the ICU have a 10% to 18% incidence of new neurologic and functional morbidities, which affect their long-term quality of life.^42–44  Thus, the developmental impacts of shifting patients from ward to ICU settings are unclear, and research efforts should be undertaken to predict and prevent hospital-acquired morbidities as this population grows.

During the study period, the use of NIV has increased more than sevenfold, as seen in Fig 3A. However, the rise in the use of NIV does not fully explain the rise in ICU burden. The absolute number of children admitted to the ICU without International Classification of Diseases (ICD) codes for IMV or NIV has also doubled. Because coding practices vary by hospital, it is likely that some children with HFNC therapy in this cohort were classified as receiving NIV, whereas others did not receive a diagnostic code for ventilatory support.⁴⁵  Although this study replicated previous methods for identifying patients receiving NIV in health care databases,⁴  the absence of dedicated ICD codes for HFNC preclude more granular analyses.⁴⁵  Thus, the rise in children being admitted to the ICU for “no” ventilatory support is likely partially representative of an increase in the use of HFNC.

Interestingly, after adjustment for confounders, increased rates of ICU admission and use of NIV were not coupled with decreased rates of IMV in the current study. Although epidemiological data from Australia and New Zealand¹⁸  had suggested that HFNC may reduce ICU admissions for bronchiolitis, studies in the United States²²  and Canada²¹  have not replicated these effects. Similarly, no randomized controlled trial has directly demonstrated a reduction in ICU admission or IMV mediated by HFNC.^14–16  Thus, the preponderance of data (including the current study) suggests that HFNC does not rescue patients destined for IMV. It remains unclear why changes in ventilatory practices have been associated with opposite effects on ICU admission rates across the globe. This may reflect differences in underlying patient population, local pathogen strains, or variability in ward and ICU acuity. It may also be that HFNC meaningfully improves respiratory distress in patients with moderate bronchiolitis in a way that cannot be captured in an administrative database.

The current study is the first to report that rates of NIV are increasing in bronchiolitis using multivariable logistic regression accounting for age, race, ethnicity, sex, insurance status, and complex chronic conditions. Increased rates of NIV and ICU admission were not coupled with significant changes in hospital length of stay (Supplemental Information: Length of Stay). These models are limited by the available covariates in PHIS. It is possible that patients with bronchiolitis have become more ill over time in a way not captured in database analyses. In an effort to address whether severity of illness was changing over time, we conducted a nested subanalysis of patients admitted to our center and found no important differences in modified PELOD or PRISM scores over time.⁴¹  Although reassuring, this subanalysis included only 3965 of 203 859 (1.9%) of patients in the main analysis. However, if patients with bronchiolitis are becoming more ill over time, it would have to be occurring in a near-linear fashion to explain the trends seen in the current study, without regard for annual variability in pathogen virulence. This seems biologically implausible given previously published epidemiological data.^46–48 

In the current study, we also found that both the incidence and cost of bronchiolitis was substantially different depending on the methodology used to define the cases (Fig 1). Although percentage of admissions with a primary diagnostic code for bronchiolitis did not significantly change over the study period, there was a significant rise in coding of bronchiolitis as a secondary diagnosis (Supplemental Information: Diagnoses). This led to markedly different estimates of incidence (Supplemental Information: Incidence), and cost (Supplemental Information: Cost). The reason for the increase in coding of bronchiolitis as a secondary diagnosis remains unclear. However, similar patterns in coding for influenza and a rise in patients with acute respiratory failure being secondarily diagnosed with bronchiolitis (Supplemental Information: Diagnoses) may suggest that diagnoses are being influenced by the use of multiplex assays.^35–37 

This study has important limitations. Although the study encompasses >200 000 admissions for bronchiolitis over 10 years, only 38 hospitals are included in the cohort. Because the PHIS database predominantly contains data from large children’s hospitals,²⁶  the data may not be nationally representative. Additionally, as is the case with many database analyses, granular data regarding severity of illness are not available, limiting understanding of rationale for clinician’s choices of ventilatory support. Finally, variable coding practices regarding the use of HFNC therapy preclude analysis of the change in this specific therapy. The establishment of methods to accurately identify patients treated with HFNC on the basis of ICD coding would substantially improve future studies.

The proportion of ICU admission for patients diagnosed with bronchiolitis has doubled over the past decade, outpacing overall ICU expansion. The use of NIV has increased sevenfold over the study period, and the use of invasive ventilation has not significantly changed. Further study is needed to better understand the factors underlying these temporal patterns, cost-effectiveness, and impact on children’s long-term health.

APR-DRG

All Patient Refined Diagnosis Group

CI

confidence interval

CMS

Centers for Medicare and Medicaid Services

GDP

gross domestic product

HFNC

high-flow nasal cannula

ICD

International Classification of Diseases

ICD-10

International Classification of Diseases, 10th Revision

ICD-9

International Classification of Diseases, Ninth Revision

IMV

invasive mechanical ventilation

IQR

interquartile range

NIV

noninvasive ventilation

PELOD

Pediatric Logistic Organ Dysfunction score

PHIS

Pediatric Health Information Systems

PRISM

Pediatric Risk of Mortality score