Design
The decision problem was ‘which is the most cost-effective treatment strategy for patients presenting to UK ambulance services with ARF?’ A cost-utility economic evaluation was performed using a probabilistic decision analytic model to synthesise available evidence and compare alternative management strategies9,10. Data from ACUTE trial is used in the base case – the rationale is that it is most relevant data for paramedic led services such as the NHS even though it is an imprecise effect estimate; and the aim of the model is to evaluate existing uncertainty around the decision problem relevant to this setting. Principles for economic evaluations outlined in the National Institute for Health and Care Excellence (NICE) Guide to the Methods of Technology Appraisal were followed11. The economic perspective was the UK health service in England and Wales with only direct treatment costs included. The model used 3.5% discount rate for costs and QALYs, and employed a lifetime horizon.
Interventions
Any potentially relevant prehospital treatments that could feasibly be implemented in the UK health service for ARF were considered. However, due to the complexity of alternative forms of Non-Invasive Ventilation, only CPAP was judged as being a practicable alternative to standard oxygen practice. Interventions therefore comprised: pre-hospital CPAP provided by ambulance service clinicians and standard oxygen therapy i.e. without pre-hospital CPAP. Hospital management was assumed identical for both comparators.
Model population and setting
The population consisted of a hypothetical cohort of patients with acute respiratory failure from any cause and potentially suitable for CPAP treatment and the setting was a representative ambulance service, such as West Midland Ambulance Service (WMAS) – setting for the ACUTE feasibility trial. Although this cohort could include patients with heterogeneous aetiology for ARF, including acute cardiogenic pulmonary oedema/heart failure, chronic obstructive pulmonary disease and pneumonia, for the purposes of modelling they were treated as a single group.
Model Structure
The model structure was based on a previously published economic model, as presented in Figure 1. Patients received prehospital CPAP in the intervention group and standard care in the comparator group, and the treatment choice affected the probability of death and probability of intubation. The model assigned a baseline probability of intubation or death within 30 days for the standard care arm; and the intervention arm probabilities were estimated by applying log odds ratios (ORs) to the baseline risks. Patients who survived accrued lifetime QALYs and health care costs according to their life expectancy. Costs were also accrued through costs of intervention (i.e., out-of-hospital CPAP) and hospital treatment costs, which depended on whether the patient needed intubation. A summary of the parameters used in the model are reported in Table 1.
Analyses were performed using different effectiveness parameters. In the base case, effectiveness data from the ACUTE pilot trial were used as it is the most representative data for paramedic led services such as the NHS, even though it is an imprecise estimate. As a pragmatic trial, ACUTE used minimal exclusion criteria to select patients and CPAP was delivered using a simple disposable unit (reference O-Two) by ambulance clinicians without extended or specialist training. A scenario analysis was undertaken using effectiveness parameters a recent network meta-analysis synthesising previously published experimental data, updated with results from the ACUTE study, using identical methods to those previously reported. The other studies included in the meta-analysis used less pragmatic inclusion criteria, more complex CPAP delivery systems and involved physicians or paramedics with online physician support to deliver the interventions. From the perspective of an ambulance service based on unsupported ambulance clinicians, the ACUTE data provide a relatively imprecise estimate of effectiveness (i.e. how the intervention works in usual practice), while the meta-analysis provides a more precise estimate of efficacy (i.e. how the intervention could work in certain circumstances).
Data
Parameter estimates, distributional forms and data sources are summarized in Table 1. The baseline risk of mortality was modelled using the 30-day mortality data from control arm of the ACUTE pilot trial which reported 9 deaths (25.7%, n=35, complete case, modified intention to treat analysis set). The intubation risk, which determines whether critical care admission is required, was also modelled using the data from control arm of the ACUTE trial, which reported one intubation (3.4%, n=29, complete case, modified intention to treat analysis set).
The base case analysis used relative effectiveness results from the ACUTE pilot trial for mortality and intubation. The odds ratios (OR) for effectiveness of CPAP for reducing mortality and intubation were 1.2 (95% CI 0.4 to 3.2) and 1.8 (95% CI 0.2 to 40.1) respectively. A scenario analysis was also performed using network meta-analysis2 revised with results from the ACUTE study, using identical methods to those previously reported. The odds ratios (OR) with 95% credible intervals for reducing mortality and intubation were 0.5 (95%CI 0.2 to 1.4) and 0.4 (0.1 to 0.9) respectively12.
Lifetime QALYs for surviving patients were estimated by multiplying the life years with representative quality of life using same estimates as in the previous economic model2; both derived from the 3CPO trial19. Discounted life expectancy was estimated at 2.67 years and the mean utility value was 0.6.
The costs included in the model are for prehospital CPAP, intubation, hospitalization, and lifetime care for patients. We estimated the costs of prehospital CPAP at an ambulance service level and converted these into a cost per patient according to a 5-year depreciation period. These costs included those for initial equipment, implementation, and ongoing maintenance. This resulted in a final CPAP cost per patient ranging from £26.53 to £39.57, which was assumed to be normally distributed around the mean of £33.00 with a standard deviation of £3.30. More details about the prehospital CPAP costing is provided in Appendix 1.
The cost of intubation was estimated in the previous HTA economic model by multiplying intensive care unit costs by the average length of stay for intubation assumed to be 5 days. These costs were inflated and this resulted in a mean annual cost of £3,600 which was parameterised as a gamma distribution with an alpha of 90 and a beta of 40, after consultation with ACUTE study clinical experts.
The hospitalisation costs were estimated as weighted average costs of non-intubated patients in the ACUTE trial that received NIV in hospital (approximately 42.5% between both arms) and that did not, which corresponded to patients with NHS Reference Cost codes DZ27S (respiratory Failure without Interventions, with CC Score 11+) and patients with code DZ27P (respiratory Failure with Single Intervention, with CC Score 11+), respectively based on expert clinical input. Thus, the mean inpatient admission cost for hospitalisations was calculated as weighted average of the costs of patients with DZ27S and DZ27P, from the NHS Reference Costs for 2016-1713, resulting in a mean cost of £3,200 and represented as gamma distribution with an alpha of 80 and a beta of 40.
Lifetime costs of survivors were estimated using the annual costs and the discounted life expectancy of patients captured from the 3CPO trial19, which were inflated resulting in a mean annual cost of £6,000. In the model, this annual cost was parameterised as a gamma distribution with an alpha of 60 and a beta of 100, after discussions with ACUTE clinical experts. It was assumed that the lifetime costs were the same for all survivors, irrespective of whether they were in the standard care or prehospital CPAP arm.
Table 1. Summary of model parameters
Parameter
|
Mean
|
Distribution or 95% CI
|
Source
|
Baseline risks
|
Risk of mortality
|
0.257
|
Beta (9,26)
|
ACUTE
|
Risk of intubation
|
0.034
|
Beta (1,28)
|
ACUTE
|
OR for prehospital CPAP
|
Base case scenario (effectiveness parameters from ACUTE)
|
Log (Mortality OR)
|
0.145
|
Normal (0.145, 0.521)
|
ACUTE
|
Log (Intubation OR)
|
0.591
|
Normal (0.591, 1.403)
|
ACUTE
|
Scenario using effectiveness parameters from NMA
|
Log (Mortality OR)
|
-0.916
|
Samples
|
ACUTE, HTA meta-analysis2
|
Log (Intubation OR)
|
-1.050
|
Samples
|
ACUTE, HTA meta-analysis2
|
Life expectancy of patients
|
Lifetime years
|
2.67 years
|
Normal (2.67, 0.16)
|
3CPO trial19, Clinical opinion
|
Health related quality of life
|
Utility
|
0.6
|
Beta (640,425)
|
3CPO trial19, Clinical opinion
|
Costs (in £)
|
Prehospital CPAP
|
£33
|
Normal (£33, £3.3)
|
O-Two/SP14, WMAS, Expert opinion
|
Hospitalisation
|
£3,200
|
Gamma (80,40)
|
NHS Reference Costs13
|
Intubation
|
£3,600
|
Gamma (90, 40)
|
HCHS index15, Clinical opinion
|
Annual costs
|
£6,000
|
Gamma (60, 100)
|
3CPO trial19, HCHS index15, Clinical opinion
|
HCHS - Hospital and Community Health Services; WMAS - West Midland Ambulance Service
Cost-effectiveness analysis
The cost-effectiveness was estimated using both the incremental cost effectiveness ratios (ICER) and the net monetary benefit (NMB) approaches. The ICER is calculated as the mean incremental cost divided by the mean incremental benefits, computed by comparing to the next most effective alternative. The willingness to pay threshold (λ) is the amount of money that the decision-maker is willing to pay to gain an additional QALY16. The usual threshold for decision-making in the UK is based on information from NICE, and considered to be £20,000 per QALY as detailed in NICE HTA guidelines. This effectively means that NICE will recommend an intervention for funding if it can deliver health gain at a cost no greater than £20,000 per QALY compared to the next most effective alternative. The NMB framework transforms cost-effectiveness results to a linear scale; it is defined as the QALYs multiplied by a value for the QALYs (e.g. £20,000) minus the costs of obtaining them: NMB= (QALYs × λ) – cost. The strategy with the highest expected incremental net monetary benefit is the most cost-effective9 17.
Probabilistic sensitivity analysis
In order to account for the uncertainty in model inputs a probabilistic sensitivity analysis (PSA) was conducted using Monte Carlo simulation.9,17,18 Multiple model runs were performed, each with a random draw from every parameter’s probabilistic distribution, thus evaluating the full range of cost-effectiveness results possible given the uncertainty on model inputs9 17 18. Mean ICERs calculated from the average expected costs and effects over all model runs, were computed and compared with cost-effectiveness thresholds to inform adoption decisions. The incremental costs and of each model run were depicted graphically on a cost-effectiveness plane. A cost-effectiveness acceptability curve (CEAC), plotting a relevant range of λ values against the probability that each intervention was the most cost-effective, was also graphed to summarise the uncertainty of PSA results19.
Value of Information analysis
The population expected value of perfect information (EVPI) places an upper limit on what healthcare system should be willing to pay for additional evidence to remove decision uncertainty i.e. EVPI informs the future total value of addition research relating to a specific decision problem9 20. The population expected value of partial perfect information (EVPPI) is the value in improving the precision of estimates of parameters, or groups of parameters.
Individual level expected value of information metrics were initially calculated for both the base case and updated meta-analysis scenario analysis. EVPI for individual patients was directly calculated directly from the model PSA output using standard formulas9 and individual EVPPI values were estimated by using 2 level Monte Carlo simulation techniques21. Assumptions on ARF incidence (11,000 patients per year in England and Wales) and health technology lifespan (5 years) were used to compute population level results.