Clinical impact of implementing humidified high-flow nasal cannula on interhospital transport among children admitted to a PICU with respiratory distress: a cohort study

Background There is a limited evidence for humidified high-flow nasal cannula (HHFNC) use on inter-hospital transport. Despite this, its use during transport is increasing in children with respiratory distress worldwide. In 2015 HHFNC was implemented on a specialized pediatric retrieval team serving for Victoria. The aim of this study is to investigate the effect of the HHFNC implementation on the retrieval team on the paediatric intensive care unit (PICU) length of stay and respiratory support use. Methods We performed a cohort study using a comparative interrupted time-series approach controlling for patient and temporal covariates, and population-adjusted analysis. We studied 3022 children admitted to a PICU in Victoria with respiratory distress January 2010–December 2019. Patients were divided in pre-intervention era (2010–2014) and post-intervention era (2015–2019). Results 1006 children following interhospital transport and 2016 non-transport children were included. Median (IQR) age was 1.4 (0.7–4.5) years. Pneumonia (39.1%) and bronchiolitis (34.3%) were common. On retrieval, HHFNC was used in 5.0% (21/420) and 45.9% (269/586) in pre- and post-intervention era. In an unadjusted model, median (IQR) PICU length of stay was 2.2 (1.1–4.2) and 1.7 (0.9–3.2) days in the pre- and post-intervention era in transported children while the figures were 2.4 (1.3–4.9) and 2.1 (1.2–4.5) days in non-transport children. In the multivariable regression model, the intervention was associated with the reduced PICU length of stay (ratio 0.64, 95% confidential interval 0.49–0.83, p = 0.001) with the predicted reduction of PICU length of stay being − 10.6 h (95% confidential interval − 16.9 to − 4.3 h), and decreased respiratory support use (− 25.1 h, 95% confidential interval − 47.9 to − 2.3 h, p = 0.03). Sensitivity analyses including a model excluding less severe children showed similar results. In population-adjusted analyses, respiratory support use decreased from 4837 to 3477 person-hour per year in transported children over the study era, while the reduction was 594 (from 9553 to 8959) person-hour per year in non-transport children. With regard to the safety, there were no escalations of respiratory support mode during interhospital transport. Conclusions The implementation of HHFNC on interhospital transport was associated with the reduced PICU length of stay and respiratory support use among PICU admissions with respiratory distress. Supplementary Information The online version contains supplementary material available at 10.1186/s13054-021-03620-7.

Full list of variables and definitions Supplemental  Table 5 Trend changes of the length of stay in the intensive care unit Supplemental Table 6 Sensitivity analysis with indicator variable for each year post intervention Supplemental Table 7 Characteristics of propensity-score matched cohorts Supplemental Table 8 Pediatric population in Victoria state Supplemental Table 9 The intubation after PICU admission following interhospital transport

Supplementary Figures
Supplemental Figure 1  To simplify the final model, the age category was modified. As children aged <6 month old and those aged 6-<12 month old have similar patient-level characteristics, these age groups were combined into a age category of 0-<1 year old.
1.2 Additional outcome report As one of secondary outcomes, we decided to add the duration of invasive ventilation so as to provide more detail results.

Additional sensitivity analysis
In the comparison of patient characteristics between transported children and non-transport children, three underlying conditions disproportionately distributed (chronic encephalopathy, home-ventilation dependent, and previous paediatric intensive care unit (PICU) admission). Post-hoc analyses with regard to these covariates were performed. A sensitivity analysis using uncensored outcomes were also added to confirm that the preset censoring did not distort the result of the primary test.

Model specification
To estimate the outcome effect by the intervention of the study, we used a comparative interrupted time series (CITS) analysis as prespecified in the protocol. Compared to the interrupted time series analysis only using the intervention group, this comparative model allowed us to calculate a more robust estimate because the outcome trend change due to secular factors and temporal changes could be set off by subtracting the outcome trend change in the comparative group from one in the intervention group. Patients were divided in one-year time period, and categorized into pre-intervention era (2010-2014) and post-intervention era (2015-2019) as per protocol. The model included a time variable, exposure to the interhospital transport, post-intervention era, and interactions among these. The trends were allowed to differ in the postintervention era. This model also controlled for patient-level variables and temporal variables.

This model is specified as:
Log Y = β0 + β1*transport + β2*year + β3*year*transport + β4*intervention* transport+ β5*intervention + β6*intervention*year*transport + β7*intervention*year + ∑ =1 + ε Y=length of PICU stay (day); transport = 1 in transported children, 0 in non-transport children; year=centralized admission year as a continuous variable (i.e. calendar year -2015); intervention = 1 in post-intervention period (2015-2019), 0 in pre-intervention era (2010-2014); λ=coefficient of covariates; X= study covariates (age, sex, cause of respiratory distress, haemato-oncological disease, neuromuscular 3 disease, airway disease, lung disease, chromosomal abnormality, chronic encephalopathy, prematurity, home-ventilation dependent, previous PICU admission, and temporal variables (HHFNC use in PICU, emergency department, and ward)) In this model, the point estimate of β4 was interpreted as the outcome effect by the intervention. We used the same standard deviation in the outcome of interest for both transported and non-transport children. The plausibility of the same standard deviation was examined in the regression diagnostics. A multivariable linear regression model with a log-transformed outcome was used according to the preset selection method for regression models based on the distribution of observed outcomes and model fitting (Supplemental figure 2, 3 and table 3). Following components with regard to the model assumption of the CITS analysis were reviewed to ascertain that the CITS analysis is a viable approach in the study cohort. We treated all admissions as independent observations since we assumed that the outcome effect would not be distorted much considering the limited number of PICU readmission within a short period. This assumption was examined by the model including only first PICU admission in the study period. We did not use a differencein-differences approach for the primary test as the trends in the pre-intervention era varied significantly between the two cohorts.
3 Assessment of model assumptions

Homogenous comparative group
To examine whether transported and non-transport children are comparative enough to be included in the CITS analysis, patient-level characteristics which could be potential confounders were compared by a descriptive data analysis as prespecified. (Supplemental table 2) Although non-transport children were more likely to have underlying diseases such as chronic encephalopathy and previous PICU admission than transported children, overall other variables were comparable. Hence, we considered that there was no major heterogeneity suggesting against using the CITS approach. The post-hoc analysis with regard to these covariates (chronic encephalopathy, home-ventilation dependent, and previous PICU admission) was added.

Linear trend
Based on literature review, the chronological trend of the length of PICU stay was linear (or can be transformed to be linear) in many studies although the direction of the trend varied across studies. In the study data, we graphically reviewed the outcome trend in non-transport children, and confirmed that there were no evidences against the analysis plan. (Supplemental figure 3)

Constant composition
Comparisons by pre-and post-intervention era in each group was listed in the table 1 of the main manuscript. Although there was an increasing trend in home-ventilation dependent and previous PICU 4 admission over year, most of patient-associated variables were comparable between pre-and post-intervention era.

Timely enforcement of the intervention
The timely increase in the HHFNC use on interhospital transport following the implementation was confirmed.
(Supplemental figure 5) 3   a Study design variables included admission year, transport, post-intervention era, an interaction between transport and post-intervention era, an interaction between year and transport, an interaction between year and post-intervention era, and an interaction between year, post-intervention era and transport.   The ratio per year (with 95% confidence interval) was presented.
Supplemental Table 6