Cost‐effectiveness of nasal high‐flow in children with acute hypoxaemic respiratory failure

A pilot randomised controlled trial assessed the early application of nasal high‐flow (NHF) therapy compared with standard oxygen therapy (SOT), in children aged 0 to 16 years presenting to paediatric emergency departments with acute hypoxaemic respiratory failure (AHRF). The study estimated the need to escalate therapy and hospital length of stay in the NHF group compared with SOT. This sub‐study then assessed the subsequent cost‐effectiveness.

Acute hypoxaemic respiratory failure (AHRF) in children aged 1-4 years is the common endpoint for many underlying specific diagnoses such as pneumonia, pneumonitis, acute lower respiratory tract infection, reactive airway disease (asthma) and bronchiolitis (in children older than 12 months of age).2][3] While childhood mortality due to respiratory infections has decreased in high-income countries, AHRF is the most frequent cause of hospital admission resulting in a major consumption of healthcare resources. 4,5In view of the global burden of respiratory disease, the World Health Organization (WHO) is advocating to develop low cost and low technology oxygen delivery methods that can be provided in health care settings. 6Currently, standard oxygen therapy (SOT) is delivered via nasal cannula with low flow oxygen up to 2 L/min.For children requiring oxygen flow between 4 and 8 L/min, a face mask is used.Excessive use of oxygen has been cautioned due to its toxicity caused by a mix of the concentration and length of exposure. 7,8A recent study in a resource limited setting in Africa, compared standard-oxygen with highflow therapy in children with pneumonia and AHRF aged between 28 days and 12 year. 2This study was terminated early due to feasibility.At the time of termination there was a trend towards a reduced 48-h mortality in children on high-flow therapy.An interesting additional finding was that NHF therapy used less oxygen resources than SOT, an important outcome for countries with limited access to oxygen.
Indeed, delivery of nasal high flow (NHF) oxygen support therapy in Emergency Departments (EDs) has now become increasingly popular as an alternative oxygen treatment to SOT.The new technology allows the inspired gas to be heated and humidified, improving pulmonary conductance and compliance compared to dry, cooler gas which allows for higher flow rates to be administered. 9,10The ease of NHF administration, improved patient comfort and little co-operation required from the child is driving the rising popularity of this treatment option. 11conomic evaluation alongside clinical trials comparing SOT with NHF therapy when used as primary support for AHRF has not yet been conducted.Clinical outcomes of the RCT applying NHF or SOT in this cohort of children aged 0 to 16 has recently been published 12 in a pilot phase of the Paediatric Acute Respiratory Intervention Study (PARIS II pilot).That study found that a large proportion of children on SOT, that met escalation criteria, were rescued with high flow and could be cared for in paediatric wards.This sub-study now reports a pre-planned economic evaluation based on data collected alongside the pilot trial, which aims to compare the cost and cost-effectiveness of providing these therapies for children with AHRF.In this sub-study, we set out to analyse the cost-effectiveness of early NHF therapy compared with SOT from a health care perspective; this will help inform decisions on clinical practice and healthcare resource allocation.

Trial design and study population and outcomes
An open-labelled, multi-centre and randomised controlled trial feasibility design was used in two tertiary children's hospitals in the ED and general wards in Australia.In total, 563 children aged between 0 and 16 years with respiratory symptoms presenting to the ED's were recruited, consented and enrolled.The inclusion criteria were children presenting with respiratory failure with an ongoing oxygen requirement to maintain oxygen saturation (SpO 2 ) of ≥92%, and admission to hospital.Patients were stratified using a pragmatic point-of-care definition into obstructive airway disease (those with wheeze) and non-obstructive airway disease (those without wheeze).The study excluded children that previously received NHF therapy during their current illness period (defined as within less than 7 days), had upper airway obstruction or a craniofacial malformation, were critically ill requiring immediate higher level of care with invasive or noninvasive ventilation, had a basal skull fracture or head trauma, had a cyanotic heart disease, were on home oxygen therapy, suffered from cystic fibrosis, received palliative care, and/or had an oncology condition.
In the NHF group (n = 283), children were to receive heated and humidified NHF delivered by the AIRVO system at weight specific rates. 12Children randomised to SOT (n = 280) were allocated to receive oxygen via nasal cannula up to 2 L/min.Children in the standard-oxygen group could receive rescue NHF therapy in the general ward settings if their condition met criteria for treatment failure.Further escalation of the non-responder SOT group to intensive care followed if the patient deteriorated while on rescue NHF in the general ward setting.For NHF non-responders, escalation was to PICU and for SOT, escalation involved NHF in the ward setting.The main characteristics of the trial design are summarised in Fig. 1.
The primary outcome of the trial was treatment failure, defined as at least three of four clinical criteria were met and escalation of care was required: (i) heart rate remained unchanged or increased compared to admission/enrolment observations; (ii) respiratory rate remained unchanged or increased compared to admission/enrolment observations; (iii) oxygen requirement on high-flow exceeded FiO2 ≥40% to maintain SpO2 ≥92% or oxygen requirement in standard-oxygen arm exceeded >2 L/min or >8 L/min to maintain SpO2 ≥92%; (iv) hospital early warning tool necessitates a medical review and the clinician decided on an escalation of care. 12he trial was registered prospectively on the Australian New Zealand Clinical Trials Registry (ACTRN12621001678886).The cost-effectiveness analysis was conducted from the healthcare service perspective.At the time of commencement, out of 563 paediatric patients, there were 15 patients who met the inclusion criteria but did not commence either therapy because it was not suitable, as decided by their treating clinician.The health economics study hence focussed on the patients in the per protocol population of the RCTs (563 paediatric patients), that is, who met inclusion criteria and were treated according to the assigned oxygen mode, with the intention-to-treat sample presented in a sensitivity analysis.

Estimating resource use and costs
The resources were costed to include those used to deliver NHF or SOT, for example, cost of hospital stay included use of ward and/or intensive care facilities for each event observed in the trial.Costs were measured from the healthcare service perspective in 2017 Australian Dollars (A$), as this was the year where a majority of data collection occurred.Given the short time horizon from the recently completed and published pilot trial and the relevant costs are already in present value terms, costs were not discounted. 13,14The time horizon over which costs and consequences are evaluated was the hospital length of stay (HLoS), calculated as the point (date/time) of randomisation to the point (date/time) of hospital discharge.
The costs of care included the inpatient ward costs, costs at ED, capital equipment/device and consumables in each intervention group, and the inpatient costs of paediatric intensive care units (PICU).Interhospital transfer costs were not assigned as there were only internal transfers to PICU.Inpatient costs at the two enrolling institutions were obtained from the administrative hospital costing units.The relevant costs included in this Cost Effectiveness Analysis (CEA) were: (i) NHF and SOT capital equipment/device, consumables in each intervention group, (ii) direct costs of medical care including medications, health care provider time and (iii) HLoS.The average ED cost per presentation was obtained from the National Hospital Cost Data Collection, Queensland, for the financial year 2017-2018 (Table 1).Length of stay in PICU was defined as date/time admitted to ICU to date/ time discharged from PICU.The cost of PICU was calculated from the total cost of PICU stay, including nursing, medical and allied health staff salary and wages, food services and cost per bed.
The cost of hospital care was calculated using the sum of the cost per ward stay and PICU stay if it occurred.Per day ward costs included nursing, medical and allied health staff salary and wages, food services, and cost per bed.

Intervention costs
The cost of each individual use of reusable NHF device (Airvo2) was based on the assumption that the total number of paediatric patients would be the same as that of patients experiencing the trial, which is 262 paediatric patients using the device each year, over a 5-year life span, while accounting for a 20% depreciation from year two.The other components of the NHF system are single-use consumables for each patient, that is, nasal cannula, oxygen tubing and wiggle pads as shown in Table 2. Cannula cost is the only consumable expense incurred for SOT treatment arm.In the cases where patients escalated from SOT to NHF therapy, the equipment and consumable cost of both SOT and NHF were included in the model.

Effectiveness of outcomes
Effectiveness was measured by HLoS in days, including all time from ED presentation to discharge from hospital.These are the records of the HLoS (days) from admission to hospital (time of randomisation) to the time of discharge.There are no missing data for HLoS.

Decision-tree economic evaluation model
To assess cost-effectiveness, a decision tree economic evaluation model was built.A simple evolution of health states over time was applied, where the patients were in one of three states: treated, escalated to PICU or discharged from hospital.No adverse events that lead to death were observed.Instead, patients experienced recovery over a short time of less than 5 days, and all were eventually discharged.

Incremental cost-effectiveness ratios
Cost effectiveness is measured using the incremental costeffectiveness ratio (ICER), calculated as cost for one additional unit of effectiveness, as follows: ICER is calculated by dividing the difference in mean costs between study groups by the difference in mean effects.
Where C NHF is the arithmetic mean cost for treatment group (Nasal High Flow); C SOT is the arithmetic mean cost comparator group (Standard Oxygen Therapy); E NHF is the arithmetic mean 2.46 † The cost of each individual use of the reusable NHF device (Airvo2) was based on 262 paediatric patients using the device each year, over a five-year life span, while accounting for a 20% depreciation from year two.Source of equipment costs: NHF costings as per Manufacturer (FPH -Fisher & Paykel Healthcare), New Zealand based on Queensland Health costs for Queensland, Australia.SOT costings as per Manufacturer (Boscomed) based on Queensland Health costs for Queensland, Australia.effect for treatment group (Nasal High Flow); E SOT is the arithmetic mean effect for comparator group (Standard Oxygen Therapy).
The ICER is estimated and interpreted as the additional (change) cost required to avoid an additional day (change) of HLoS.It can be presented visually on a cost-effectiveness plane, a graph with 2 axestypically costs on the y-axis and effects on the x-axis, as illustrated in Fig. 2. A point ICER estimate is first obtained and then the ICERs from the bootstrapped sample (explained later) are plotted on the cost-effectiveness plane as a 'cloud' of points representing uncertainty 15 and cost-effectiveness acceptability curves are used to show the probability that the new intervention (i.e., nasal high flow) should be considered cost-effective at a range of willingness-to-pay values. 16he ICERS are measured in two mutually exclusive diagnoses (obstructiveongoing bronchodilator and/or steroid therapy vs. nonobstructivepneumonia, pneumonitis, acute lower respiratory tract infection) compared with two mutually exclusive outcomes (responder -first-line treatment success vs. non-responder -first-line treatment failure).Full details are provided in Table 1A (Supporting Information) Hence, there were four groups of analysis: (i) obstructive responder, (ii) obstructive non-responder, (iii) non-obstructive responder and (iv) non-obstructive non-responder.
Using the total costs and HLoS for both the treatment and control groups the ICER values were computed for responders and non-responders in each classification (obstructive and nonobstructive), hence we calculated four ICERs, so that our study accounts for the underlying differences in acuity across diagnostic groups (obstructive vs. non-obstructive).The ICER for each outcome represents the incremental cost of using NHF, compared to SOT, for the change in HLoS for children, in days.
Where confidence levels for parameter values were reported in the pilot trial, these were used to inform the distribution and ranges for both deterministic and probabilistic sensitivity analyses.
After assigning the appropriate distribution for each parameter, a non-parametric bootstrapping method was completed to describe the distribution of mean values and compare the costs and outcomes between groups.Bootstrapping then enabled presentation of stochastic data into a graph. 18The values of each parameter were randomly selected and non-parametric bootstrapping (1000 replications) were used to estimate uncertainty using 95% credible intervals around the point estimates and the probability that treatment NHF was effective, compared to the SOT.Again, this was conducted across all four mutually exclusive patient clinical outcomes compared between SOT and NHF: obstructive and responder; obstructive and non-responder; non obstructive and responder; non obstructive and non-responder.

Sensitivity analysis
A sensitivity analysis was performed to test the methodologic assumptions and sampling variability for the intention-to-treat sample (583 paediatric patients) with the results presented using cost-effectiveness acceptability curves.The bootstrapped results were presented in cost-effective planes, and each of the 1000 dots represents one mean cost and effectiveness outcome from a bootstrapped sample with replacement of the original sample. 18ootstrap methods are used to enable inference about a population from sample data, by resampling from the sample data and.It is assumed that the empirical distribution of the data is an adequate representation of the true distribution of the data.The statistical analysis is therefore based on repeatedly sampling from the observed data.In health economic evaluation, it is popular to use bootstrapped resamples with replacement to provide ICERs for hypothetical repeats of the study, with the assumption that the study sample is a valid representation of the underlying population of interest.The nonparametric bootstrap has been recommended as the primary statistical test for making inferences about arithmetic means for small to moderately sized samples of highly skewed data, 18 which fits for use in our study.
Analyses were conducted and reported following the costeffectiveness analysis alongside clinical trials good practice guideline 19 and the Consolidated Health Economic Evaluation Reporting Standards reporting guideline. 20The Stata statistical software package (version 14.0) (Stat Corp, College Station, TX) and Microsoft Excel was used.

Resource use
As noted earlier, all children in the per-protocol population were included in the trial.Therefore, of the 548 actually enrolled into the study, 262 received NHF and 286 received SOT.Baseline characteristics of the children were similar between the two allocated intervention groups. 12Escalation of care occurred in 50 children (17.5%) in the SOT arm and in 32 children (12.2%) in the NHF arm.All 50 children in the SOT were escalated to NHF in the general ward and a further 20 children required PICU admission with the remaining 30 successfully treated with NHF in the general ward.All 32 children in the NHF arm who required escalation of care were transferred to PICU as no alternative rescue therapy could be offered on the wards.Fig. 2 Cost-effectiveness plane. 17he costs are separated in the obstructive and non-obstructive groups, with the group of tables (Tables 3-6) showing the relevant data of mean cost, effectiveness outcomes, for both SOT and NHF group within each strata.Since we measure the ICER for each group, this is interpreted as the additional cost per change in hospital length of stay in hospital.In 3 of the sub groups, there is an increase in length of stay, that is, a negative change in length of stay.The costs increase also, so there is a negative change in costs.The point estimate is therefore in the NW quadrant and deemed not cost-effectivethis is the same when we use the bootstrap estimates for the three sub-groups.In the final sub-group, when the change in length of stay is positive (i.e., a lower length of stay for NHF), and the change in costs remain negative also, the ICER is then in the NE quadrant and NHF is deemed cost-effective if below a threshold of willingness-to-pay.Using the bootstrap estimates, we show that overall there is costeffectiveness of NHF for that group.For example, an ICER of À$440.86 means that NHF is cost-effective if health services are willing to spend an additional A$440.86 per one unit of hospital length of stay avoided.
For the SOT and NHF in the obstructive responder group (Table 3), the mean costs of inpatient stay were significantly lower for the SOT A$2005.07 versus the NHF A$2559.22 with a difference of A$554.16 (95% CI A$76.37,A$1031.94).The mean effectiveness outcome (HLoS) was also significantly higher for the SOT 1.77 days versus the NHF 3.02 days with a difference of 1.26 days (95% CI 0.04, 2.49), indicating that children in the SOT treatment arm group had shorter hospital length of stay compared to those in the NHF group.Similar patterns could be observed in the other two groups: obstructive non-responder (Table 4), non-obstructive responder (Table 5).However, difference in effectiveness outcome was not significant in these two groups.For the SOT and NHF in the non-obstructive nonresponder group (Table 6), the mean effectiveness outcome was non-significantly lower for the SOT 6.32 days versus the NHF 5.21 days with a difference of 1.11 days (95% CI À4.45 to 2.23) while the mean costs of inpatient stay were significantly lower for the SOT A$13 188.29 versus the NHF A$13 594.73 with a difference of A$406.45 (95% CI A$À10 255.19,A$9442.30).
For children who were diagnosed as an obstructive responder to treatment, the ICER result demonstrated that NHF would be cost-effective if health services were willing to spend A$440.86 to lower hospital length of stay by 1 day.For children who were obstructive responders to treatment, NHF was cost-effective if A$52 167.76 is spent to reduce one-day hospital length of stay.The reasoning for this difference is as follows: at the patient-level costing, patients in this sub-group randomised to NHF and failed this treatment arm all were escalated to ICU (non-responder), meanwhile only one-third of the control arm (SOT) were escalated to ICU.Moreover, we can notice that the effect difference (HLoS) between these two arms was almost trivial (3.83 vs. 3.95), being the least significant difference among four subgroups.Since ICER is based on change of cost and effect, in this sub-group, the large cost change divided by the small effect change contributed to this much larger ICER.
Children who presented as a non-obstructive responder to treatment had an ICER result showing NHF was cost-effective if ) † Data are presented as mean cost per patient (SD), expressed as A$.Including administrator, medical, nurse salaries, consumables and equipment to process the admission and transfer of the patient from the emergency department to the ward and discharge from hospital.‡ Average hospital length of stay (SD), in days.A positive difference in hospital length of stay indicates a larger number of days in patients who received the SOT intervention.A negative difference indicates more days in hospital for those who received NHF treatment.To be deemed cost effective, an ICER depends on the amount health services are willing to pay threshold to avoid a further night in hospital.À0.12 (À0.88 to 1.11) † Data are presented as mean cost per patient (SD), expressed as A$.Including administrator, medical, nurse salaries, consumables and equipment to process the admission and transfer of the patient from the emergency department to the ward and discharge from hospital.‡ Average hospital length of stay (SD), in days.A positive difference in hospital length of stay indicates a larger number of days in patients who received the SOT intervention.A negative difference indicates more days in hospital for those who received NHF treatment.To be deemed cost effective, an ICER depends on the amount health services are willing to pay threshold to avoid a further night in hospital.
health services were willing to spend A$469.56 per one unit of reduced length of stay (one less night in hospital).Conversely, those who presented as non-obstructive non-responder to treatment had an ICER result that showed NHF to be cost-effective if health services were willing to spend A$367.20 per one night less in hospital, demonstrating cost-effectiveness for this group.

Probabilistic sensitivity analysis
To account for sampling uncertainty, probabilistic sensitivity analysis was undertaken using bootstrapping methods as outline earlier.All results and interpretation of the 1000 simulations are presented in the cost-effectiveness plane graphs (Figs 3-6).Each

ObstrucƟve Responder
Fig. 3 The cost-effectiveness plane for the obstructive responder group.Including administrator, medical, nurse salaries, consumables and equipment to process the admission and transfer of the patient from the emergency department to the ward and discharge from hospital.‡ Average hospital length of stay (SD), in days.A positive difference in hospital length of stay indicates a larger number of days in patients who received the SOT intervention.A negative difference indicates more days in hospital for those who received NHF treatment.To be deemed cost effective, an ICER depends on the amount health services are willing to pay threshold to avoid a further night in hospital..77(À0.15 to 1.70) † Data are presented as mean cost per patient (SD), expressed as A$.Including administrator, medical, nurse salaries, consumables and equipment to process the admission and transfer of the patient from the emergency department to the ward and discharge from hospital.‡ Average hospital length of stay (SD), in days.A positive difference in hospital length of stay indicates a larger number of days in patients who received the SOT intervention.A negative difference indicates more days in hospital for those who received NHF treatment.To be deemed cost effective, an ICER depends on the amount health services are willing to pay threshold to avoid a further night in hospital.
dot represents one mean cost and effectiveness comparison from a random sampling of replacement of the original sample.
This figure shows cost effectiveness plane for obstructive responder group and that all estimates fall on the north-west quadrant (more costly, less effective), so there is no indication of cost-effectiveness.
This demonstrates that 75% of bootstrapped estimates fall on north-east (more costly, more effective), 24.2% of them fall on north-west (more costly, less effective).As the majority of estimates are cost-effective, the treatment (NHF) could be cost-effective but this depends on the willingness to pay threshold.
This plane shows that the treatment (NHF) costs more and patients have more hospital stay days.It shows that an overwhelming estimate (94.6%) falls on north-west quadrant (more costly, less effective).
Figure 5 shows cost-effectiveness for the non-obstructive nonresponder group and that 51.1% of bootstrapped values fall on the south-east (more effective, less costly) and 32.8% of them fall on the north-east quadrant (more effective, more costly).
The sensitivity analysis is presented in Figures 1A-4A (Supporting Information), and the results are quite consistent with the primary analysis for the intention-to-treat scenario.In this scenario, NHF is not cost-effective in either the obstructive responder or non-obstructive responder groups.In the obstructive non-responder group, it is likely that NHF is more effective but more costly (Quadrant 1) compared to SOT, with 75% of bootstrapped values in this quadrant versus 41.2% in primary analysis.

Incremental Cost
Incremental Effect ObstrucƟve Non-responder Fig. 4 The cost-effectiveness plane for the obstructive non-responder group.
-$300.00  less costly' (Quadrant 2) in the sensitivity analysis compared to 37% of those values in the primary analysis.
Next, the non-obstructive non-responder is explored further to ascertain the extent of willingness-to-pay for each treatment at different levels of cost-effectiveness.The probability of NHF to be cost-effective is 76% at the value for money threshold of A$10 000 per hospital day, for the nonobstructive non-responder group.From this threshold onwards, the probability remains almost the same (Fig. 7).The sensitivity analysis demonstrates similar results (Fig. 5A, Supporting Information).

Discussion
While childhood mortality due to respiratory infections has decreased in high-income countries, AHRF is the most frequent cause of hospital admission in children resulting in major consumption of healthcare resources. 4,5In informing resource allocation and clinical practice, evidence taking into account both the cost and the effectiveness of treatment arms is needed.This study compared the cost-effectiveness of NHF and SOT, and found that NHF is not cost-effective for children who are responsive to their first line of treatment (no escalation), irrespective of their diagnosis of obstructive or non-obstructive airway disease.][23] While NHF has been shown to be safe and improve the work of breathing, the length of stay in hospital is longer with an additional increased cost for more expensive oxygenation equipment.In this secondary analysis of the parent RCT, it was then investigated whether there is a specific subgroup of children with AHRF that may benefit from NHF therapy in relation to health care costs.

Obstructive responder group
The finding demonstrates that the NHF intervention was not cost-effective for the obstructive responder group.The ICER result in this group demonstrates that NHF was more costly and less effective in terms of HLoS, located in the fourth quadrant of the cost-effectiveness plane.Interestingly, when the bootstrap method was applied to the 1000-replication data set, all estimates fell into the NW quadrant of the cost-effectiveness plane.This indicates that NHF is both less effective and will also increase cost.It can be inferred that NHF is not cost effective compared to SOT when a patient is obstructive and responsive to their first treatment option.
Obstructive non-responder group NHF intervention was likely cost-effective; however, health services may have to be willing to spend A$52 167.76 to reduce length of stay by one unit.75% of estimates fell in the NE quadrant, indicating that 75% of NHF treatment cases were likely to observe increased cost with increased effectiveness.This indicates that for children who are classified as obstructive and nonresponder to the first line of treatment, they were likely to be severe cases of AHRF and resource intensive.Indeed, these cases may require escalation of therapy beyond the scope of NHF or SOT treatment.

Non-obstructive and responder group
Similar to the obstructive responder category, the ICER results for this group was low, at A$469.56 per one unit of additional length of stay.Our bootstrap results demonstrated that 94.6% of respondents fell into the NW plane, that is, more costly, less effective.NHF was not cost effective compared to SOT when patients are non-obstructive and responder to first line of treatment.

Non-obstructive and non-responder group
This group has the lowest ICER value of A$367.20 per one unit of length of stay.In comparison with the other groups, this was the lowest value that health providers would face when deciding whether to implement NHF.Once bootstrapped, 84% of estimates are more effective, aggregating 32.8% more costly and 51.1% less.The remaining 16% were less effective and largely equally distributed between being more costly and less costly.

Implications for practice
Studies have found that NHF benefits children with bronchiolitis, with escalation of care being lower in children receiving NHF compared to SOT, however no cost benefit for NHF could be demonstrated. 12,23,24The parent RCT to this analysis showed a similar benefit of NHF therapy in children with acute respiratory failure.This economic evaluation of the parent RCT went one step further and observes how these benefits translate to costeffectiveness between children who were responsive to their first line of treatment and those who were non-responder.The results demonstrate that NHF is not cost-effective for children who respond to their first line of treatment (no escalation), irrespective of their diagnosis of obstructive or non-obstructive airway disease.Conversely, when children required escalation in care (non-responder), NHF were observed to be cost-effective in some cases.From these findings, it can be inferred that the NHF treatment option was cost-effective for the proportion of children requiring escalation of care under NHF.
To ensure applicability beyond the per-protocol setting, sensitivity analyses with the intention-to-treat sample was undertaken, to capture other potential gaps.Considering the generalisability of the cost-effectiveness findings, the findings of the sensitivity analyses supported the key conclusions.

Strengths and limitations
A key strength is that this study included paediatric patients who were randomly assigned to groups, hence minimising the bias of self-selection.The use of patient-level cost data from the hospital admission to discharge also lends towards precision of cost estimates.
A common limitation of within trial economic evaluation is the short time horizon. 19In addition, the cost-effectiveness analysis results were drawn upon from the point when children were discharged from hospital and longer effects with relevant costs could not be measured nor taken into account.Cost-effectiveness analysis also may consider a societal perspective to provide additional information on out-of-pocket expenses, and also could collect information on patient quality of life.Surveys were developed for patient level data, to add to the trial data, but the data collection process in ED proved difficult and was suspended.
While children in the study were appropriately grouped according to presenting illness, the assumption of a homogenous history outside of their ED presentation is a limitation.Morbidity, or a previous history of AHRF may intrinsically bias the results.Taking these factors into account would make for a complicated model and would be an appropriate future consideration.
Further, this study accounted for equipment and consumable usage and costs for children based on the study protocol.The assumption was made that children who were randomised to NHF will use the consumables and equipment once, as are the children allocated to SOT for their respective apparatus.It does not account for instances where children require more than one set of reusable consumables.This has been noted to occur when the original set does not fit properly and so another larger set must be opened; movement between units may result in equipment left behind; or damage to existing sets.This was not tracked in the data and so we are unable to comment on how this may have affected our cost-effectiveness estimates, but given that the larger costs are hospital stays, the overall impact would be minimal.

Conclusion
In times of scarce health resources there is a need to build an investment case for new health technologies such as NHF.Economic evaluations support decision makers to allocate health resources effectively by maximising the health gain.In contribution to the literature, this study has examined the costeffectiveness of NHF compared with SOT.As first-line treatment, NHF is unlikely to be cost-effective compared with SOT in obstructive responder, obstructive non-responder, and non-obstructive responder group in either per-protocol or intentionto-treat sample.But, for non-obstructive patients who required escalation in care (non-obstructive non-responder), NHF is likely to be cost-effective if willingness to pay per reduced hospital length of stay is more than A$10 000 per patient.This study highlights the value of a cost-effectiveness analysis in the allocation of resources for developing new models of care.

Fig. 1
Fig.1Main characteristics of the trial design.12

Table 2
Unit costs of intervention equipment

Table 1
Hospital costs associated with hospital length of stay Standard Error) presented.Routine patient-level cost data in Australian hospitals consists of hospitals costs associated with the inpatient admission.Detailed cost categories such as imaging, pathology, nursing, medical, pharmacy, allied services, and paediatric intensive care unit (PICU) are included.

Table 3
Cost and effectiveness outcomes in the obstructive responder group

Table 4
Cost and effectiveness outcomes in the obstructive non-responder group

Table 6
Cost and effectiveness outcomes in the non-obstructive non-responder group 23) † Data are presented as mean cost per patient (SD), expressed as A$.

Table 5
Cost and effectiveness outcomes in the non-obstructive responder group