Early Use of Anti-influenza Medications in Hospitalized Children With Tracheostomy

We conducted a multicenter, retrospective cohort study using the Pediatric Health Information System (PHIS) database. The PHIS database contains administrative and billing data from inpatient and observation units, ED, and ambulatory surgery encounters at 49 freestanding pediatric hospitals in the United States. Data are deidentified at the time of submission and subjected to rigorous quality checks before inclusion in the database.²⁴ The database allows longitudinal patient tracking by using encrypted patient medical record numbers. The Children’s Hospital Los Angeles Institutional Review Board reviewed the study and granted an exemption according to 45 Code of Federal Regulations 46.101(b)(4).

We included patients with admission age 30 days to 19 years discharged between October 1, 2007, and September 30, 2015, with (1) International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes representing presence of tracheostomy (V44.0, V55.0, 519.00, 519.00, 519.01, 519.02, and 519.09) and (2) influenza infection (487.xx and 488.xx; Fig 1). These ICD-9-CM codes have been previously used to identify hospitalized pediatric patients with the presence of tracheostomy^19,21,25 and laboratory-confirmed influenza with high specificity and positive predictive value.^26,–29 

We excluded patients with any of the following characteristics: (1) age 20 years or older to exclude adult patients; (2) transfer from an outside hospital with inpatient status, given interest in initial treatment of influenza and length of stay (LOS); (3) extreme outliers in LOS defined as median plus 3 times the interquartile range (IQR); (4) death during the index hospitalization because the patients could not have a revisit; (5) presence of surgical procedures unrelated to influenza within first 2 days of hospitalization to minimize the possibility of including patients who acquired or manifested with influenza infection during hospitalizations for other reasons; or (6) missing variables in the database. To prevent overrepresentation by patients with multiple eligible encounters, we first assigned a computer-generated random whole number to each admission episode and subsequently selected 1 admission episode for each patient for analysis.

Our primary exposure was early receipt of anti-influenza medications, defined as prescription of amantadine, rimantadine, oseltamivir, zanamivir, or peramivir on day 0 or 1 of hospitalization. We used this definition to be inclusive and minimize bias, considering our patient recruitment period. Given the robust data quality management and encryption of medical record numbers and other identifying patient information, we did not review clinical charts to validate whether data around anti-influenza medication administration reported in PHIS were accurate.

Our primary outcomes were LOS in days and 30-day all-cause revisit, defined as any repeat presentation to an inpatient unit, observation unit, or ED. Previous studies have used LOS^28,30 and 30-day all-cause revisit²⁹ to measure the impact of anti-influenza therapy in children who were hospitalized. We chose to evaluate all-cause revisits rather than those related to influenza or other respiratory tract infections, given the wide variety of comorbidities of patients with tracheostomy and difficulty assessing causality of revisit based on administrative data alone. Our secondary outcome was the number of days the patient received antibiotics that treat bacterial tracheostomy-associated respiratory tract infections.³¹ 

Demographic covariates included age at admission in years, sex, race and/or ethnicity, and insurance status. Race and/or ethnicity was categorized as non-Hispanic white, non-Hispanic African American, Hispanic, and other. Insurance status was categorized as private, government, and other. Medical covariates were complex chronic conditions (CCCs), presence of gastrostomy, baseline dependence of mechanical ventilation, receipt of care at ICU, and antibiotics on day 0 or 1 of hospitalization. CCCs have previously been defined on the basis of specific ICD-9-CM codes.^32,33 Gastrostomy status was defined as having any of the ICD-9-CM codes at discharge: v44.1, v55.1, or 536.4x.³⁴ Baseline ventilator dependence was defined as having any of the ICD-9-CM codes falling under v46.1x at discharge.²¹ 

For all discharge episodes that met inclusion and exclusion criteria, descriptive statistics (eg, mean, SD, and frequencies) were calculated for the entire sample and for both exposure groups. Next, bivariate analyses were conducted to determine if receipt of early anti-influenza medications was associated with any of the covariates and outcomes of interest. We used χ² tests for categorical covariates and Student’s t test for continuous covariates.

To assess the effect of receipt of early anti-influenza medication on our outcomes, we subsequently used propensity score matching to adjust for confounding by indication.³⁵ We included the following covariates (chosen a priori) to generate propensity score on the basis of multivariate logistic regression: patient age, sex, race and/or ethnicity, presence of neurologic and neuromuscular CCCs, premature birth, total number of CCCs, presence of gastrostomy tube, baseline ventilator dependence, and receipt of systemic antibiotics and ICU care on day 0 or 1 of hospitalization. Propensity score and matched samples were generated in R (version 3.5.0) by using the package MatchIt.^36,37 We performed 1-to-1 matching without replacement using optimal matching strategy, aiming to minimize total within-pair difference of the propensity score (optimal matching). Patients without matches were removed from analysis. We used Student’s t test to compare LOS and days on antibiotics. χ² test was used for comparison of 30-day all-cause revisit rate.

A total of 1436 discharge episodes met inclusion criteria of which 899 discharge episodes were eligible for propensity score matching (Fig 1). Characteristics of the cohort eligible for propensity score matching are described in Table 1. The median age of the cohort was 5 years (IQR: 2–10 years). The cohort was predominately male (56.5%; n = 508), non-Hispanic white (40.4%; n = 363), and on public insurance (75.5%; n = 679). The median number of CCCs was 3 (IQR: 3–4), and all patients had respiratory CCCs. Gastrointestinal (82.9%; n = 745), neurologic and neuromuscular (50.6%; n = 466), and other congenital CCCs (33.5%; n = 314) were also commonly identified. Most patients had technology dependence, such as gastrostomytube (73.6%; n = 676) and baseline mechanical ventilator dependence (25.5%; n = 229). Nearly 40% received care at ICU after admission. Systemic antibiotics were prescribed to the majority of the patients after admission (n = 739; 82.2%) and for most days of hospitalization (average 5.3 days; 95% confidence interval [CI] 5.0–5.6).

Of this non–propensity-matched cohort, 513 patients (57.1%) received anti-influenza medications on hospital day 0 or 1. In this group, 499 received oseltamivir only, 10 received both oseltamivir and calcium channel blockers (amantadine or rimantadine), and only 2 received calcium channel blockers. No patient in our study cohort received peramivir or zanamivir. Those who received anti-influenza medications were older (median age of 5 vs 4 years; P = .004) and more likely to have gastrointestinal CCCs (85.2% vs 79.8%; P = .03). Those receiving anti-influenza medications had statistically shorter LOS (average 6.6 vs 7.5 days; P = .02) but no differences in 30-day revisit rates (26.1% vs 24.1%; P = .49) or days of antibiotics (5.2 vs 5.6 days; P = .19). Variations at the hospital level accounted for between 0% and 2.3% of variance in the outcomes of interest.

We successfully matched 386 patients who received early anti-influenza medications to those who did not receive early anti-influenza medications. Characteristics of this population are detailed in Table 2. After matching, the differences in age and gastrointestinal CCCs were not statistically significant. Of the 386 that received early anti-influenza medications, 379 (98.2%) received oseltamivir as the sole anti-influenza medication on day 0 or 1. As noted in Table 3, the LOS was shorter for the cohort that received early anti-influenza medications (6.4 vs 7.5 days; P = .01), representing an absolute change in LOS of 1.1 days (95% CI 0.3–2.0). No statistically significant differences were detected in 30-day all-cause revisit rate (27.5% for early anti-influenza medications versus 24.1% for no early anti-influenza medications; P = .28). After a closer look at revisit rates, rates of hospital readmission and of ED visits within 30 days of discharges were similar in the 2 cohorts. Systemic antibiotics were prescribed on the majority of hospital days in both cohorts with no statistically significant difference (5.0 vs 5.6 days; P = .11).

Because we excluded 127 patients in the matched propensity analysis, we conducted a post hoc sensitivity analysis to assess if the outcomes and effect sizes seen in the matched cohort remained stable when those who were unmatched were included. To perform the sensitivity analyses, we conducted traditional multivariable regression (linear for LOS and logistic for 30-day revisit) of all 899 eligible patients. We included all demographic and clinical characteristics in Table 1 as covariates. We obtained results like those seen in the propensity score–matched cohort for which receipt of early anti-influenza medications was independently associated with shorter LOS (absolute difference −0.98 days; P = .01) and similar 30-day revisit rates (adjusted odds ratio 1.12; P = .48).

In our retrospective, propensity score–matched cohort study of nearly 800 children who were hospitalized with tracheostomy and influenza, we have demonstrated that early administration of anti-influenza medications is associated with shorter LOS by >1 day and no increase in 30-day all-cause revisit rate. Our study population was notable for a high degree of medical complexity with multiple CCCs and technology dependence. More than 80% of our study population received systemic antibiotics after admission. Regardless of receipt of early anti-influenza medications, systemic antibiotics were given for >75% of hospital days in our population with documented viral infection.

Although early administration of anti-influenza medication decreases influenza-related complications for CMC in the outpatient setting,^13,14 this study is the first to reveal that early anti-influenza medication administration may improve clinical outcomes in a subpopulation of CMC. The results of our study are comparable to previous studies that revealed association with anti-influenza medications and shorter LOS without increase in readmission rate among children who experienced ICU admissions²⁸ or complications from influenza.²⁹ The changes in LOS in the current study are similar to the 1-day reduction in duration of illness for children who receive anti-influenza medications in the outpatient setting.^14,38 Compared with the population in these studies, our population was notable for high medical complexity with multiple chronic care conditions. Our study captured a unique population that was not previously studied, and early use of anti-influenza medications was again associated with similar improvements in LOS.

The rate of timely administration of anti-influenza medications was <60% in our study population. This indicates that, despite recommendations, clinicians have difficulty following the clinical guidelines for early anti-influenza medication administration^16,39 in this medically complex patient population that has increased morbidity from influenza and its complications. With our results, we affirm the recommendation for timely administration of antiviral medications in all children hospitalized with influenza³⁹ by demonstrating its impact in a population of CMC. The LOS reductions seen in the current study may decrease family and caregiver burden and costs to the health care system. Previous quality improvement efforts to ensure the provision of early anti-influenza medications in children who were hospitalized uncovered barriers for providers to adhere to the guideline recommendations.⁴⁰ Given the impact revealed in our study, future quality improvement efforts might specifically target CMC to ensure the provision of early anti-influenza medication administration.

We found widespread and prolonged use of systemic antibiotics in our patient cohort despite identification of a viral infection. There are several possible reasons for high rates of prolonged antibiotic use in our study. First, there are no evidence-based guidelines for differentiating and diagnosing bacterial and viral tracheostomy-associated respiratory tract infections. With high rates of positive bacterial respiratory culture results in children with tracheostomy,³¹ some providers may continue to treat positive respiratory culture results in the setting of a known viral infection. Second, bacterial coinfection is a known complication of influenza^6,41,42; therefore, some providers may choose to continue antibiotics because of concerns for bacterial coinfection or secondary complications. Third, some patients may have had concomitant bacterial infections (eg, urinary tract infections) for which they appropriately continued to receive antibiotics. Fourth, concomitant infections with other respiratory viruses may have contributed to delayed response to anti-influenza medications, and clinicians may have chosen to continue antibiotics. Finally, some clinicians may be reluctant to discontinue antibiotics on a patient who is clinically improving on antibiotics, even with a viral source. The role of continued antibiotics in viral tracheostomy-related respiratory tract infections may merit future research.

Our study has some limitations because of the retrospective design using an administrative database. First, because the PHIS database only includes freestanding children’s hospitals, it is possible that exclusion of patients transferred from inpatient units of other hospitals resulted in exclusion of children with higher acuity; thus, our results may not be generalizable to all patients with tracheostomy hospitalized with influenza. Second, the PHIS database has no laboratory test results; therefore, we may have missed some patients with influenza or inappropriately included patients who did not have documented influenza. However, in previous studies, ICD-9-CM codes have been shown to have high specificity and positive predictive value for identifying children hospitalized with laboratory-confirmed influenza.^26,27 Third, although we attempted to adjust for confounding by indication by using propensity score matching, additional important patient and hospital characteristics, such as influenza vaccination status and duration of illness before admission, are not available from the database. These patient-level and hospital-level factors could not be included in our propensity score but may have influenced anti-influenza medication prescribing patterns and study outcomes. The impact of these factors on propensity for timely receipt of anti-influenza medications and outcomes in children who are hospitalized is incompletely understood. A previous study of >30 000 admissions due to influenza (including >6000 children) revealed similar rates of timely administration of anti-influenza medications regardless of duration of illness preceding hospitalization.¹⁷ Differences in care pathways between individual hospitals were unavailable from the database and may have accounted for differences in our outcomes. However, we found that differences in treating hospitals accounted for 0% to 2.3% of variance in all outcomes, which suggests any variations due to hospital-level factors, such as care pathways, did not have significant influence on the outcomes. Lastly, because mortality is rare among children hospitalized with influenza,^6,–8 our cohort lacked sufficient patient deaths to examine this as an outcome.

The strengths of this study are its large sample size and use of propensity score matching. Propensity score matching is a statistical method used to adjust for confounding by indication,³⁵ and it has been used in studies in which authors use PHIS data to assess the effect of various interventions.^28,34,43 We chose this method rather than multivariate regression because of its potential to handle a larger number of covariates with the sample size of our study⁴⁴ and to simulate randomized controlled trials by distinguishing covariates, exposure, and outcomes a priori.³⁵ Although excluding those without matches could introduce bias, our sensitivity analysis using multivariate regression analyses with all eligible subjects found similar results. Because of ethical challenges conducting randomized, placebo-controlled prospective trials with anti-influenza medications,⁴⁵ quasi-experimental design such as this study may provide the strongest evidence for early use of anti-influenza medication.

Early administration of anti-influenza medications is associated with reduction of LOS by 15% in children who are hospitalized with tracheostomy. With our results, we support the recommendation for timely administration of these medications for all children hospitalized with influenza.^16,39 With ongoing difficulties in providing effective influenza prevention^11,12 and increased risk of hospitalization,^4,6,–8 adherence to current influenza treatment guidelines in the inpatient setting remains especially important for the care of CMC. With this study, we also offer insight into antibiotic prescribing behavior regarding children with tracheostomy hospitalized with viral respiratory tract infections. Authors of future studies should investigate the role of quality improvement efforts in optimizing the care of CMC with influenza and the effect of antibiotics in children with tracheostomy during viral illness.

CCC

complex chronic condition

CI

confidence interval

CMC

children with medical complexity

ED

emergency department

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

IQR

interquartile range

LOS

length of stay

PHIS

Pediatric Health Information System