Day of Illness and Outcomes in Bronchiolitis Hospitalizations

This investigation was a secondary analysis from 2 multicenter prospective studies in which follow-up strategies after bronchiolitis hospitalizations in generally healthy infants and children <2 years of age were examined. The first was an observational study of patients hospitalized at 5 hospitals (including 3 community hospitals) between December 2016 and March 2017.⁹  The second, the Bronchiolitis Follow-up Intervention Trial, was a randomized controlled trial (RCT) comparing follow-up strategies after hospital discharge, conducted at 4 hospitals (including 2 community hospitals) between January 2018 to April 2018 and November 2018 to April 2019.⁸ 

Depicted in Table 1 is a breakdown of the contribution of patients from each study. For both studies, patients were considered for inclusion if they were hospitalized with an attending physician diagnosis of bronchiolitis and did not have complex medical conditions, including congenital heart disease, chronic lung disease, neuromuscular disease, or immunodeficiencies. Although the participating institutions encourage use of American Academy of Pediatrics guidelines²  for admission and discharge criteria, adherence to these guidelines was not mandatory for study inclusion. Patients were excluded for family refusal to participate and, in the RCT, if the primary team or primary care provider felt follow-up was indicated. Patients discharged from the hospital on supplemental oxygen were excluded from both original studies; however, because outcome data were collected on these patients in the RCT, we opted to include these patients for this analysis. Additionally, we collected LOS and PPV for all excluded patients in the RCT and included these patients in the analysis of LOS and PPV, although we did not collect information on PPV in the observational study. Total cough duration was available for patients in both studies who completed phone follow-up.

The predictor variable was day of illness at the time of admission to the hospital. Investigators extracted this information from the admission history and physical and/or emergency department (ED) or urgent care notes. Day of illness was calculated as the total elapsed time, in days, between the date that the first symptom (eg, fever, runny nose, cough, difficulty feeding, difficulty breathing) related to bronchiolitis started and the date of admission. If notes described an interval with complete resolution of symptoms before admission, then the earlier period of symptoms was not included in the day of illness calculation.

The outcome variables included LOS, receipt of PPV, and total cough duration. We defined PPV as the need for endotracheal intubation or continuous or bilevel positive airway pressure, but we also examined a definition of PPV that included the use of high flow nasal cannula (HFNC). The study team extracted outcome variables from the medical record, with the exception of cough duration, which was obtained in both investigations via weekly phone follow-up calls conducted until the cough was reported by the caregiver to have stopped. We analyzed the association between day of illness and total cough duration from the time of admission but also report descriptive data on cough duration from the day of hospital discharge.

We conducted multivariable linear and Poisson regression, as appropriate, to analyze the association between the predictor and outcome variables, with adjustment for age, sex, insurance (government versus nongovernment), race, and ethnicity. We used mixed effects models to adjust for clustering by hospital. Comorbidities, including previous wheezing, eczema, and prematurity, were assessed during in-hospital interviews with caregivers on enrolled but not excluded patients, so these variables were added to the multivariable model as a sensitivity analysis. We did not include virology results as a covariate because most patients who underwent viral testing only had limited testing (for respiratory syncytial virus and influenza), but we did conduct a post hoc sensitivity analysis for respiratory syncytial virus–positive patients

Severity of illness might confound any association between day of illness and outcomes (ie, practitioners might factor day of illness into their decision to admit to the hospital). To assess for a potential association between day of illness and severity of illness, we adapted a scoring tool used by Freire et al¹⁰  to predict escalated care in bronchiolitis in the ED setting, and we applied this tool to a random sample of 100 patients. Please refer to the Supplemental Information for further explanation of our application of this tool. Because there was no association between severity of illness and day of illness (correlation coefficient: −0.01, P = .91), we opted not to include severity of illness as a covariate.

Because the sample size was predetermined for the 2 studies, we did not conduct an a priori sample size calculation; precision of our estimates is reflected in their confidence intervals (CIs).

Day of illness and LOS data were available for 746 patients. The median (interquartile range [IQR]) age was 6.4 (2.6–13) months, 423 (56.7%) patients were male, 318 (42.6%) had government insurance, 514 (68.9%) were identified as white race, and 229 (30.7%) were identified as Hispanic ethnicity (Table 2). The median (IQR) day of illness at admission was 4 (2–5) days, with 198 of 746 (26.5%) patients presenting before the third day of illness and with 121 of 746 (16.2%) presenting after the fifth day (Fig 1). The median (IQR) LOS was 2 (1–3) days, a total of 218 of 746 (29.2%) patients were hospitalized in the ICU, and for the 548 patients for whom the degree of respiratory support was documented, 88 (16.1%) received PPV and 233 (42.5%) received PPV or HFNC.

The unadjusted association between day of illness at admission and LOS is depicted in Fig 2, with day of illness categorized into 5 groups. On multivariable analysis, day of illness at admission was not associated with LOS (coefficient for each additional day of illness on presentation 0.01 days, 95% CI: −0.05 to 0.08 days, P = .68), PPV (incidence rate ratio [IRR] for each additional day of illness on presentation 1.0, 95% CI: 0.9 to 1.1, P = .94), or total cough duration (coefficient 0.33 days, 95% CI: −0.01 to 0.67 days, P = .06) (Table 3). When HFNC is included in the definition of PPV, the association with day of illness is also not statistically significant (IRR 0.95, 95% CI: 0.9 to 1, P = .13). The addition to the model of previous wheezing, eczema, and prematurity for the 563 patients on whom those variables were available had minimal impact on effect sizes and did not impact statistical significance for any of the 3 outcomes (Table 3). A post hoc analysis of patients who were RSV positive similarly did not impact effect sizes or statistical significance.

Information on hospital readmission was available for 424 patients. A total of 16 (3.8%) patients were readmitted, and there was no significant difference in the day of illness at discharge in readmitted versus nonreadmitted patients (5.9 vs 6.4 days, P = .83 with adjustment for age, sex, insurance, race, and ethnicity). Patients discharged early in the course of their illness (third day of illness or earlier) had a similar readmission risk to those discharged on or beyond their fourth day of illness (3.4% vs 3.8%, P = .87). The median (IQR) cough duration after hospital discharge was 6 (3–10) days, with 65 (14.3%) patients having cough for 2 weeks or more.

In this secondary analysis of 2 prospective, multicenter studies involving hospitalized bronchiolitis patients, we observed no association between day of illness at the time of admission and subsequent outcomes, including LOS, receipt of PPV, or total duration of cough. Our findings should help inform the anticipatory guidance provided to families during bronchiolitis hospitalizations and might discourage practitioners from using day of illness to influence clinical decision-making.

A commonly used educational resource for clinical guidance states that bronchiolitis “begins with upper respiratory tract symptoms, followed by lower respiratory tract signs and symptoms on days 2 to 3, which peak on days 3 to 5 and then gradually resolve.”⁵  Authors of 1 bronchiolitis review article describe day of illness as “an important variable in providing anticipatory guidance for outpatient management.”⁶  We are not aware of any data describing how frequently practitioners factor in the day of illness into their clinical decision-making. However, in our review of admission notes for these 2 studies, we noted that most assessments and plans included mention of day of illness, inferring that this variable is often part of the patient’s narrative.

Although researchers of multiple studies have investigated total symptom duration in infants and young children with bronchiolitis, we were unable to find reliable evidence supporting a peak in severity between 3 and 5 days in either the inpatient or outpatient setting. Swingler et al¹¹  demonstrated that the median duration of illness in outpatients evaluated for bronchiolitis was 12 days, with 18% still reporting symptoms after 21 days, whereas 25% of patients with bronchiolitis seen in an ED were noted to have persistent symptoms after 21 days.¹²  However, researchers of neither study evaluated the timing of peak symptoms or evolution of severity of illness. Leidy et al¹³  and Robbins et al¹⁴  described prolonged recoveries after bronchiolitis hospitalizations but similarly did not describe the time course of disease severity. Authors of some studies of bronchiolitis therapies have reported days of symptoms before enrollment and daily illness severity after administration of the study drug¹⁵  but do not aggregate these time points at the patient level, which makes deciphering the exact peak in severity challenging. Researchers of human challenge studies involving young, healthy adults have demonstrated that symptoms tend to correlate with viral load, tend to start within 1 to 2 days of inoculation, and peak ∼5 to 8 days after inoculation.^16,17 

Therefore, the notion that disease severity tends to peak on days 3 to 5 may be derived from adult experimental models but is not necessarily supported or contradicted by evidence from children. Although diseases that are acute and self-limited by definition will exhibit a peak in severity sometime between symptom onset and symptom resolution, with our findings, we suggest that severe bronchiolitis (ie, that which requires hospitalization) may be a heterogeneous disease with a variable and unpredictable trajectory. Recent data describing viral coinfections in ∼30% of hospitalized patients may be used to explain some of this heterogeneity,¹⁸  although associations between being infected with >1 virus and hospital LOS were not particularly strong in that study.

Rather than using the day of illness to anchor clinical management and anticipatory guidance, day-to-day progress may be a more reliable predictor of the disease trajectory. In a prospective multicenter study of 1916 patients hospitalized with bronchiolitis, Mansbach et al¹⁹  demonstrated that deterioration after a period of clinical improvement was rare, occurring in only 4% of patients. The authors also noted that patients with ≥1 day of breathing difficulty at presentation had no difference when compared with patients with <1 day of breathing difficulty in the likelihood of subsequent clinical worsening after having initial improvement, which supports our finding that day of illness has poor predictive value.

However, in a related study from the same patient cohort, the authors noted that patients with <1 day breathing difficulty at the time of presentation had higher odds of needing mechanical ventilation (odds ratio of 1.6, 95% CI: 1.2 to 2.1), a finding that motivated our analysis of PPV.¹⁸  The authors of that study drew comparisons between their findings and rapidly progressive influenza, where deaths have been reported to occur early in the course of the illness.^20,21  However, their findings contrast with ours, even when HFNC is added to the definition of PPV. Although this contrast may be explained by differences in study design and sample size, it suggests that associations between the timing of the disease course and the need for escalation of respiratory support remain unclear.

One potential explanation for the lack of association between day of illness and outcomes in our study is that severity of illness at presentation may confound this association. For example, practitioners may be more likely to hospitalize an infant who is early in the course of their illness because of fear that they may worsen. This phenomenon would bias toward the null hypothesis because patients earlier in the course would therefore be less ill and expected to fare better, thereby neutralizing any association with day of illness. Using the score developed by Freire et al,¹⁰  we found that severity of illness in a random sample of 100 patients was not associated with day of illness, which gave us some reassurance that severity of illness was not an important confounder in our study.

We did not attempt to adjust for severity of illness at discharge. Day of illness might also impact the decision to discharge, but in this case, any confounding due to severity of illness would bias away from the null hypothesis in that patients earlier in the course of illness would be hospitalized longer although they are less ill. Therefore, this potential bias should not have influenced our main findings.

In our study, we provide descriptive data on duration of cough after discharge. Our finding that cough persists for about a week in most patients but ≥2 weeks in 14.3% (1 in 7) of patients may be useful as anticipatory guidance at discharge. An understanding that symptoms are often persistent may prevent unneeded subsequent visits.

Our study has several limitations. First, because of differences in protocols for the original studies, the data fields differed, which affected the sample sizes for some outcomes (eg, we did not record PPV data in the observational study). Second, we did not have precise criteria for determination of onset of illness and relied on documentation from admission notes. Although similar approaches have been used in multiple other bronchiolitis studies to determine illness duration on admission,^15,22–24  this measure is subject to caregiver recall and may not reflect the true trajectory of the disease. Third, total cough duration was assessed by weekly phone calls. Although parents were instructed on enrollment that they would be asked about duration of symptoms, symptom diaries were not required. Fourth, we did not have comprehensive virology data because most patients who underwent viral testing were only tested for RSV and influenza. In future investigations, researchers should further evaluate the influence of viral etiology on disease trajectory. Finally, for PPV, we did not record the date of initiation, which might have provided additional insight into the timing of peak severity of illness.

Caregiver-reported day of illness at presentation does not appear to be associated with hospital LOS, receipt of PPV, or total cough duration in bronchiolitis hospitalizations. Severe bronchiolitis is a heterogeneous condition with an unpredictable clinical course. Practitioners should exercise caution when making clinical decisions or providing anticipatory guidance on the basis of symptom duration.

CI

confidence interval

ED

emergency department

HFNC

high flow nasal cannula

IQR

interquartile range

IRR

incidence rate ratio

LOS

length of stay

PPV

positive-pressure ventilation

RCT

randomized controlled trial